llvm::InstCombiner Class Reference

InstCombiner - The -instcombine pass. More...

#include <InstCombine.h>

Inheritance diagram for llvm::InstCombiner:

Public Types
typedef IRBuilder< true, TargetFolder, InstCombineIRInserter >	BuilderTy

Public Member Functions
	InstCombiner ()
bool	runOnFunction (Function &F) override
bool	DoOneIteration (Function &F, unsigned ItNum)
void	getAnalysisUsage (AnalysisUsage &AU) const override
const DataLayout *	getDataLayout () const
TargetLibraryInfo *	getTargetLibraryInfo () const
Instruction *	visitAdd (BinaryOperator &I)
Instruction *	visitFAdd (BinaryOperator &I)
Value *	OptimizePointerDifference (Value *LHS, Value *RHS, Type *Ty)
Instruction *	visitSub (BinaryOperator &I)
Instruction *	visitFSub (BinaryOperator &I)
Instruction *	visitMul (BinaryOperator &I)
Value *	foldFMulConst (Instruction *FMulOrDiv, Constant *C, Instruction *InsertBefore)
Instruction *	visitFMul (BinaryOperator &I)
Instruction *	visitURem (BinaryOperator &I)
Instruction *	visitSRem (BinaryOperator &I)
Instruction *	visitFRem (BinaryOperator &I)
bool	SimplifyDivRemOfSelect (BinaryOperator &I)
Instruction *	commonRemTransforms (BinaryOperator &I)
Instruction *	commonIRemTransforms (BinaryOperator &I)
	Common integer remainder transforms. More...
Instruction *	commonDivTransforms (BinaryOperator &I)
Instruction *	commonIDivTransforms (BinaryOperator &I)
	Common integer divide transforms. More...
Instruction *	visitUDiv (BinaryOperator &I)
Instruction *	visitSDiv (BinaryOperator &I)
Instruction *	visitFDiv (BinaryOperator &I)
Value *	FoldAndOfICmps (ICmpInst *LHS, ICmpInst *RHS)
	FoldAndOfICmps - Fold (icmp)&(icmp) if possible. More...
Value *	FoldAndOfFCmps (FCmpInst *LHS, FCmpInst *RHS)
Instruction *	visitAnd (BinaryOperator &I)
Value *	FoldOrOfICmps (ICmpInst *LHS, ICmpInst *RHS)
	FoldOrOfICmps - Fold (icmp)|(icmp) if possible. More...
Value *	FoldOrOfFCmps (FCmpInst *LHS, FCmpInst *RHS)
Instruction *	FoldOrWithConstants (BinaryOperator &I, Value *Op, Value *A, Value *B, Value *C)
Instruction *	visitOr (BinaryOperator &I)
Instruction *	visitXor (BinaryOperator &I)
Instruction *	visitShl (BinaryOperator &I)
Instruction *	visitAShr (BinaryOperator &I)
Instruction *	visitLShr (BinaryOperator &I)
Instruction *	commonShiftTransforms (BinaryOperator &I)
Instruction *	FoldFCmp_IntToFP_Cst (FCmpInst &I, Instruction *LHSI, Constant *RHSC)
Instruction *	FoldCmpLoadFromIndexedGlobal (GetElementPtrInst *GEP, GlobalVariable *GV, CmpInst &ICI, ConstantInt *AndCst=nullptr)
Instruction *	visitFCmpInst (FCmpInst &I)
Instruction *	visitICmpInst (ICmpInst &I)
Instruction *	visitICmpInstWithCastAndCast (ICmpInst &ICI)
Instruction *	visitICmpInstWithInstAndIntCst (ICmpInst &ICI, Instruction *LHS, ConstantInt *RHS)
Instruction *	FoldICmpDivCst (ICmpInst &ICI, BinaryOperator *DivI, ConstantInt *DivRHS)
Instruction *	FoldICmpShrCst (ICmpInst &ICI, BinaryOperator *DivI, ConstantInt *DivRHS)
	FoldICmpShrCst - Handle "icmp(([al]shr X, cst1), cst2)". More...
Instruction *	FoldICmpAddOpCst (Instruction &ICI, Value *X, ConstantInt *CI, ICmpInst::Predicate Pred)
	FoldICmpAddOpCst - Fold "icmp pred (X+CI), X". More...
Instruction *	FoldGEPICmp (GEPOperator *GEPLHS, Value *RHS, ICmpInst::Predicate Cond, Instruction &I)
Instruction *	FoldShiftByConstant (Value *Op0, Constant *Op1, BinaryOperator &I)
Instruction *	commonCastTransforms (CastInst &CI)
	Implement the transforms common to all CastInst visitors. More...
Instruction *	commonPointerCastTransforms (CastInst &CI)
	Implement the transforms for cast of pointer (bitcast/ptrtoint) More...
Instruction *	visitTrunc (TruncInst &CI)
Instruction *	visitZExt (ZExtInst &CI)
Instruction *	visitSExt (SExtInst &CI)
Instruction *	visitFPTrunc (FPTruncInst &CI)
Instruction *	visitFPExt (CastInst &CI)
Instruction *	visitFPToUI (FPToUIInst &FI)
Instruction *	visitFPToSI (FPToSIInst &FI)
Instruction *	visitUIToFP (CastInst &CI)
Instruction *	visitSIToFP (CastInst &CI)
Instruction *	visitPtrToInt (PtrToIntInst &CI)
Instruction *	visitIntToPtr (IntToPtrInst &CI)
Instruction *	visitBitCast (BitCastInst &CI)
Instruction *	visitAddrSpaceCast (AddrSpaceCastInst &CI)
Instruction *	FoldSelectOpOp (SelectInst &SI, Instruction *TI, Instruction *FI)
Instruction *	FoldSelectIntoOp (SelectInst &SI, Value *, Value *)
Instruction *	FoldSPFofSPF (Instruction *Inner, SelectPatternFlavor SPF1, Value *A, Value *B, Instruction &Outer, SelectPatternFlavor SPF2, Value *C)
Instruction *	visitSelectInst (SelectInst &SI)
Instruction *	visitSelectInstWithICmp (SelectInst &SI, ICmpInst *ICI)
Instruction *	visitCallInst (CallInst &CI)
Instruction *	visitInvokeInst (InvokeInst &II)
Instruction *	SliceUpIllegalIntegerPHI (PHINode &PN)
Instruction *	visitPHINode (PHINode &PN)
Instruction *	visitGetElementPtrInst (GetElementPtrInst &GEP)
Instruction *	visitAllocaInst (AllocaInst &AI)
Instruction *	visitAllocSite (Instruction &FI)
Instruction *	visitFree (CallInst &FI)
Instruction *	visitLoadInst (LoadInst &LI)
Instruction *	visitStoreInst (StoreInst &SI)
Instruction *	visitBranchInst (BranchInst &BI)
Instruction *	visitSwitchInst (SwitchInst &SI)
Instruction *	visitInsertValueInst (InsertValueInst &IV)
Instruction *	visitInsertElementInst (InsertElementInst &IE)
Instruction *	visitExtractElementInst (ExtractElementInst &EI)
Instruction *	visitShuffleVectorInst (ShuffleVectorInst &SVI)
Instruction *	visitExtractValueInst (ExtractValueInst &EV)
Instruction *	visitLandingPadInst (LandingPadInst &LI)
Instruction *	visitInstruction (Instruction &I)
Instruction *	InsertNewInstBefore (Instruction *New, Instruction &Old)
Instruction *	InsertNewInstWith (Instruction *New, Instruction &Old)
Instruction *	ReplaceInstUsesWith (Instruction &I, Value *V)
Instruction *	EraseInstFromFunction (Instruction &I)
void	computeKnownBits (Value *V, APInt &KnownZero, APInt &KnownOne, unsigned Depth=0) const
bool	MaskedValueIsZero (Value *V, const APInt &Mask, unsigned Depth=0) const
unsigned	ComputeNumSignBits (Value *Op, unsigned Depth=0) const
Pass *	createPrinterPass (raw_ostream &O, const std::string &Banner) const override
	createPrinterPass - Get a function printer pass. More...
void	assignPassManager (PMStack &PMS, PassManagerType T) override
PassManagerType	getPotentialPassManagerType () const override
	Return what kind of Pass Manager can manage this pass. More...
PassKind	getPassKind () const
virtual const char *	getPassName () const
AnalysisID	getPassID () const
	getPassID - Return the PassID number that corresponds to this pass. More...
virtual bool	doInitialization (Module &)
virtual bool	doFinalization (Module &)
virtual void	print (raw_ostream &O, const Module *M) const
void	dump () const
virtual void	preparePassManager (PMStack &)
	Check if available pass managers are suitable for this pass or not. More...
void	setResolver (AnalysisResolver *AR)
AnalysisResolver *	getResolver () const
virtual void	releaseMemory ()
virtual void *	getAdjustedAnalysisPointer (AnalysisID ID)
virtual ImmutablePass *	getAsImmutablePass ()
virtual PMDataManager *	getAsPMDataManager ()
virtual void	verifyAnalysis () const
virtual void	dumpPassStructure (unsigned Offset=0)
template<typename AnalysisType >
AnalysisType *	getAnalysisIfAvailable () const
bool	mustPreserveAnalysisID (char &AID) const
template<typename AnalysisType >
AnalysisType &	getAnalysis () const
template<typename AnalysisType >
AnalysisType &	getAnalysis (Function &F)
template<typename AnalysisType >
AnalysisType &	getAnalysisID (AnalysisID PI) const
template<typename AnalysisType >
AnalysisType &	getAnalysisID (AnalysisID PI, Function &F)
void	visit (Iterator Start, Iterator End)
void	visit (Module &M)
void	visit (Function &F)
void	visit (BasicBlock &BB)
void	visit (Module *M)
void	visit (Function *F)
void	visit (BasicBlock *BB)
Instruction *	visit (Instruction *I)
Instruction *	visit (Instruction &I)
void	visitModule (Module &M)
void	visitFunction (Function &F)
void	visitBasicBlock (BasicBlock &BB)
Instruction *	visitReturnInst (ReturnInst &I)
Instruction *	visitIndirectBrInst (IndirectBrInst &I)
Instruction *	visitResumeInst (ResumeInst &I)
Instruction *	visitUnreachableInst (UnreachableInst &I)
Instruction *	visitAtomicCmpXchgInst (AtomicCmpXchgInst &I)
Instruction *	visitAtomicRMWInst (AtomicRMWInst &I)
Instruction *	visitFenceInst (FenceInst &I)
Instruction *	visitTruncInst (TruncInst &I)
Instruction *	visitZExtInst (ZExtInst &I)
Instruction *	visitSExtInst (SExtInst &I)
Instruction *	visitFPTruncInst (FPTruncInst &I)
Instruction *	visitFPExtInst (FPExtInst &I)
Instruction *	visitFPToUIInst (FPToUIInst &I)
Instruction *	visitFPToSIInst (FPToSIInst &I)
Instruction *	visitUIToFPInst (UIToFPInst &I)
Instruction *	visitSIToFPInst (SIToFPInst &I)
Instruction *	visitPtrToIntInst (PtrToIntInst &I)
Instruction *	visitIntToPtrInst (IntToPtrInst &I)
Instruction *	visitBitCastInst (BitCastInst &I)
Instruction *	visitAddrSpaceCastInst (AddrSpaceCastInst &I)
Instruction *	visitVAArgInst (VAArgInst &I)
Instruction *	visitDbgDeclareInst (DbgDeclareInst &I)
Instruction *	visitDbgValueInst (DbgValueInst &I)
Instruction *	visitDbgInfoIntrinsic (DbgInfoIntrinsic &I)
Instruction *	visitMemSetInst (MemSetInst &I)
Instruction *	visitMemCpyInst (MemCpyInst &I)
Instruction *	visitMemMoveInst (MemMoveInst &I)
Instruction *	visitMemTransferInst (MemTransferInst &I)
Instruction *	visitMemIntrinsic (MemIntrinsic &I)
Instruction *	visitVAStartInst (VAStartInst &I)
Instruction *	visitVAEndInst (VAEndInst &I)
Instruction *	visitVACopyInst (VACopyInst &I)
Instruction *	visitIntrinsicInst (IntrinsicInst &I)
Instruction *	visitCastInst (CastInst &I)
Instruction *	visitBinaryOperator (BinaryOperator &I)
Instruction *	visitCmpInst (CmpInst &I)
Instruction *	visitTerminatorInst (TerminatorInst &I)
Instruction *	visitUnaryInstruction (UnaryInstruction &I)

Static Public Member Functions
static const PassInfo *	lookupPassInfo (const void *TI)
static const PassInfo *	lookupPassInfo (StringRef Arg)
static Pass *	createPass (AnalysisID ID)

Data Fields
InstCombineWorklist	Worklist
	Worklist - All of the instructions that need to be simplified. More...
BuilderTy *	Builder

Static Public Attributes
static char	ID = 0

Protected Member Functions
bool	skipOptnoneFunction (const Function &F) const

Private Member Functions
bool	ShouldChangeType (Type *From, Type *To) const
Value *	dyn_castNegVal (Value *V) const
Value *	dyn_castFNegVal (Value *V, bool NoSignedZero=false) const
Type *	FindElementAtOffset (Type *PtrTy, int64_t Offset, SmallVectorImpl< Value * > &NewIndices)
Instruction *	FoldOpIntoSelect (Instruction &Op, SelectInst *SI)
bool	ShouldOptimizeCast (Instruction::CastOps opcode, const Value *V, Type *Ty)
Instruction *	visitCallSite (CallSite CS)
Instruction *	tryOptimizeCall (CallInst *CI, const DataLayout *DL)
bool	transformConstExprCastCall (CallSite CS)
Instruction *	transformCallThroughTrampoline (CallSite CS, IntrinsicInst *Tramp)
Instruction *	transformZExtICmp (ICmpInst *ICI, Instruction &CI, bool DoXform=true)
Instruction *	transformSExtICmp (ICmpInst *ICI, Instruction &CI)
bool	WillNotOverflowSignedAdd (Value *LHS, Value *RHS)
bool	WillNotOverflowUnsignedAdd (Value *LHS, Value *RHS)
Value *	EmitGEPOffset (User *GEP)
Instruction *	scalarizePHI (ExtractElementInst &EI, PHINode *PN)
Value *	EvaluateInDifferentElementOrder (Value *V, ArrayRef< int > Mask)
bool	SimplifyAssociativeOrCommutative (BinaryOperator &I)
Value *	SimplifyUsingDistributiveLaws (BinaryOperator &I)
Value *	SimplifyDemandedUseBits (Value *V, APInt DemandedMask, APInt &KnownZero, APInt &KnownOne, unsigned Depth)
bool	SimplifyDemandedBits (Use &U, APInt DemandedMask, APInt &KnownZero, APInt &KnownOne, unsigned Depth=0)
Value *	SimplifyShrShlDemandedBits (Instruction *Lsr, Instruction *Sftl, APInt DemandedMask, APInt &KnownZero, APInt &KnownOne)
bool	SimplifyDemandedInstructionBits (Instruction &Inst)
Value *	SimplifyDemandedVectorElts (Value *V, APInt DemandedElts, APInt &UndefElts, unsigned Depth=0)
Value *	SimplifyVectorOp (BinaryOperator &Inst)
	Makes transformation of binary operation specific for vector types. More...
Instruction *	FoldOpIntoPhi (Instruction &I)
Instruction *	FoldPHIArgOpIntoPHI (PHINode &PN)
Instruction *	FoldPHIArgBinOpIntoPHI (PHINode &PN)
Instruction *	FoldPHIArgGEPIntoPHI (PHINode &PN)
Instruction *	FoldPHIArgLoadIntoPHI (PHINode &PN)
Instruction *	OptAndOp (Instruction *Op, ConstantInt *OpRHS, ConstantInt *AndRHS, BinaryOperator &TheAnd)
Value *	FoldLogicalPlusAnd (Value *LHS, Value *RHS, ConstantInt *Mask, bool isSub, Instruction &I)
Value *	InsertRangeTest (Value *V, Constant *Lo, Constant *Hi, bool isSigned, bool Inside)
Instruction *	PromoteCastOfAllocation (BitCastInst &CI, AllocaInst &AI)
Instruction *	MatchBSwap (BinaryOperator &I)
bool	SimplifyStoreAtEndOfBlock (StoreInst &SI)
Instruction *	SimplifyMemTransfer (MemIntrinsic *MI)
Instruction *	SimplifyMemSet (MemSetInst *MI)
Value *	EvaluateInDifferentType (Value *V, Type *Ty, bool isSigned)
Value *	Descale (Value *Val, APInt Scale, bool &NoSignedWrap)

Private Attributes
const DataLayout *	DL
TargetLibraryInfo *	TLI
bool	MadeIRChange
LibCallSimplifier *	Simplifier
bool	MinimizeSize

Detailed Description

InstCombiner - The -instcombine pass.

Definition at line 86 of file InstCombine.h.

Member Typedef Documentation

typedef IRBuilder<true, TargetFolder, InstCombineIRInserter> llvm::InstCombiner::BuilderTy

Builder - This is an IRBuilder that automatically inserts new instructions into the worklist when they are created.

Definition at line 101 of file InstCombine.h.

Constructor & Destructor Documentation

llvm::InstCombiner::InstCombiner ( )

inline

Definition at line 105 of file InstCombine.h.

                 : FunctionPass(ID), DL(nullptr), Builder(nullptr) {
    MinimizeSize = false;
    initializeInstCombinerPass(*PassRegistry::getPassRegistry());
  }

Member Function Documentation

void FunctionPass::assignPassManager ( PMStack &  PMS, PassManagerType  PreferredType  )

overridevirtualinherited

Find appropriate Function Pass Manager or Call Graph Pass Manager in the PM Stack and add self into that manager.

Reimplemented from llvm::Pass.

Definition at line 1865 of file LegacyPassManager.cpp.

                                                                    {

  // Find Function Pass Manager
  while (!PMS.empty()) {
    if (PMS.top()->getPassManagerType() > PMT_FunctionPassManager)
      PMS.pop();
    else
      break;
  }

  // Create new Function Pass Manager if needed.
  FPPassManager *FPP;
  if (PMS.top()->getPassManagerType() == PMT_FunctionPassManager) {
    FPP = (FPPassManager *)PMS.top();
  } else {
    assert(!PMS.empty() && "Unable to create Function Pass Manager");
    PMDataManager *PMD = PMS.top();

    // [1] Create new Function Pass Manager
    FPP = new FPPassManager();
    FPP->populateInheritedAnalysis(PMS);

    // [2] Set up new manager's top level manager
    PMTopLevelManager *TPM = PMD->getTopLevelManager();
    TPM->addIndirectPassManager(FPP);

    // [3] Assign manager to manage this new manager. This may create
    // and push new managers into PMS
    FPP->assignPassManager(PMS, PMD->getPassManagerType());

    // [4] Push new manager into PMS
    PMS.push(FPP);
  }

  // Assign FPP as the manager of this pass.
  FPP->add(this);
}

Instruction * InstCombiner::commonCastTransforms

(

CastInst &

CI

)

Implement the transforms common to all CastInst visitors.

Definition at line 294 of file InstCombineCasts.cpp.

                                                            {
  Value *Src = CI.getOperand(0);

  // Many cases of "cast of a cast" are eliminable. If it's eliminable we just
  // eliminate it now.
  if (CastInst *CSrc = dyn_cast<CastInst>(Src)) {   // A->B->C cast
    if (Instruction::CastOps opc =
        isEliminableCastPair(CSrc, CI.getOpcode(), CI.getType(), DL)) {
      // The first cast (CSrc) is eliminable so we need to fix up or replace
      // the second cast (CI). CSrc will then have a good chance of being dead.
      return CastInst::Create(opc, CSrc->getOperand(0), CI.getType());
    }
  }

  // If we are casting a select then fold the cast into the select
  if (SelectInst *SI = dyn_cast<SelectInst>(Src))
    if (Instruction *NV = FoldOpIntoSelect(CI, SI))
      return NV;

  // If we are casting a PHI then fold the cast into the PHI
  if (isa<PHINode>(Src)) {
    // We don't do this if this would create a PHI node with an illegal type if
    // it is currently legal.
    if (!Src->getType()->isIntegerTy() ||
        !CI.getType()->isIntegerTy() ||
        ShouldChangeType(CI.getType(), Src->getType()))
      if (Instruction *NV = FoldOpIntoPhi(CI))
        return NV;
  }

  return nullptr;
}

Instruction* llvm::InstCombiner::commonDivTransforms

(

BinaryOperator &

I

)

Instruction * InstCombiner::commonIDivTransforms

(

BinaryOperator &

I

)

Common integer divide transforms.

This function implements the transforms common to both integer division instructions (udiv and sdiv). It is called by the visitors to those integer division instructions.

Definition at line 713 of file InstCombineMulDivRem.cpp.

                                                                 {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  // The RHS is known non-zero.
  if (Value *V = simplifyValueKnownNonZero(I.getOperand(1), *this)) {
    I.setOperand(1, V);
    return &I;
  }

  // Handle cases involving: [su]div X, (select Cond, Y, Z)
  // This does not apply for fdiv.
  if (isa<SelectInst>(Op1) && SimplifyDivRemOfSelect(I))
    return &I;

  if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
    // (X / C1) / C2  -> X / (C1*C2)
    if (Instruction *LHS = dyn_cast<Instruction>(Op0))
      if (Instruction::BinaryOps(LHS->getOpcode()) == I.getOpcode())
        if (ConstantInt *LHSRHS = dyn_cast<ConstantInt>(LHS->getOperand(1))) {
          if (MultiplyOverflows(RHS, LHSRHS,
                                I.getOpcode() == Instruction::SDiv))
            return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
          return BinaryOperator::Create(I.getOpcode(), LHS->getOperand(0),
                                        ConstantExpr::getMul(RHS, LHSRHS));
        }

    if (!RHS->isZero()) { // avoid X udiv 0
      if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
        if (Instruction *R = FoldOpIntoSelect(I, SI))
          return R;
      if (isa<PHINode>(Op0))
        if (Instruction *NV = FoldOpIntoPhi(I))
          return NV;
    }
  }

  if (ConstantInt *One = dyn_cast<ConstantInt>(Op0)) {
    if (One->isOne() && !I.getType()->isIntegerTy(1)) {
      bool isSigned = I.getOpcode() == Instruction::SDiv;
      if (isSigned) {
        // If Op1 is 0 then it's undefined behaviour, if Op1 is 1 then the
        // result is one, if Op1 is -1 then the result is minus one, otherwise
        // it's zero.
        Value *Inc = Builder->CreateAdd(Op1, One);
        Value *Cmp = Builder->CreateICmpULT(
                         Inc, ConstantInt::get(I.getType(), 3));
        return SelectInst::Create(Cmp, Op1, ConstantInt::get(I.getType(), 0));
      } else {
        // If Op1 is 0 then it's undefined behaviour. If Op1 is 1 then the
        // result is one, otherwise it's zero.
        return new ZExtInst(Builder->CreateICmpEQ(Op1, One), I.getType());
      }
    }
  }

  // See if we can fold away this div instruction.
  if (SimplifyDemandedInstructionBits(I))
    return &I;

  // (X - (X rem Y)) / Y -> X / Y; usually originates as ((X / Y) * Y) / Y
  Value *X = nullptr, *Z = nullptr;
  if (match(Op0, m_Sub(m_Value(X), m_Value(Z)))) { // (X - Z) / Y; Y = Op1
    bool isSigned = I.getOpcode() == Instruction::SDiv;
    if ((isSigned && match(Z, m_SRem(m_Specific(X), m_Specific(Op1)))) ||
        (!isSigned && match(Z, m_URem(m_Specific(X), m_Specific(Op1)))))
      return BinaryOperator::Create(I.getOpcode(), X, Op1);
  }

  return nullptr;
}

Instruction * InstCombiner::commonIRemTransforms

(

BinaryOperator &

I

)

Common integer remainder transforms.

This function implements the transforms common to both integer remainder instructions (urem and srem). It is called by the visitors to those integer remainder instructions.

Definition at line 1194 of file InstCombineMulDivRem.cpp.

                                                                 {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  // The RHS is known non-zero.
  if (Value *V = simplifyValueKnownNonZero(I.getOperand(1), *this)) {
    I.setOperand(1, V);
    return &I;
  }

  // Handle cases involving: rem X, (select Cond, Y, Z)
  if (isa<SelectInst>(Op1) && SimplifyDivRemOfSelect(I))
    return &I;

  if (isa<Constant>(Op1)) {
    if (Instruction *Op0I = dyn_cast<Instruction>(Op0)) {
      if (SelectInst *SI = dyn_cast<SelectInst>(Op0I)) {
        if (Instruction *R = FoldOpIntoSelect(I, SI))
          return R;
      } else if (isa<PHINode>(Op0I)) {
        if (Instruction *NV = FoldOpIntoPhi(I))
          return NV;
      }

      // See if we can fold away this rem instruction.
      if (SimplifyDemandedInstructionBits(I))
        return &I;
    }
  }

  return nullptr;
}

Instruction * InstCombiner::commonPointerCastTransforms

(

CastInst &

CI

)

Implement the transforms for cast of pointer (bitcast/ptrtoint)

Definition at line 1431 of file InstCombineCasts.cpp.

                                                                   {
  Value *Src = CI.getOperand(0);

  if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Src)) {
    // If casting the result of a getelementptr instruction with no offset, turn
    // this into a cast of the original pointer!
    if (GEP->hasAllZeroIndices() &&
        // If CI is an addrspacecast and GEP changes the poiner type, merging
        // GEP into CI would undo canonicalizing addrspacecast with different
        // pointer types, causing infinite loops.
        (!isa<AddrSpaceCastInst>(CI) ||
          GEP->getType() == GEP->getPointerOperand()->getType())) {
      // Changing the cast operand is usually not a good idea but it is safe
      // here because the pointer operand is being replaced with another
      // pointer operand so the opcode doesn't need to change.
      Worklist.Add(GEP);
      CI.setOperand(0, GEP->getOperand(0));
      return &CI;
    }

    if (!DL)
      return commonCastTransforms(CI);

    // If the GEP has a single use, and the base pointer is a bitcast, and the
    // GEP computes a constant offset, see if we can convert these three
    // instructions into fewer.  This typically happens with unions and other
    // non-type-safe code.
    unsigned AS = GEP->getPointerAddressSpace();
    unsigned OffsetBits = DL->getPointerSizeInBits(AS);
    APInt Offset(OffsetBits, 0);
    BitCastInst *BCI = dyn_cast<BitCastInst>(GEP->getOperand(0));
    if (GEP->hasOneUse() &&
        BCI &&
        GEP->accumulateConstantOffset(*DL, Offset)) {
      // Get the base pointer input of the bitcast, and the type it points to.
      Value *OrigBase = BCI->getOperand(0);
      SmallVector<Value*, 8> NewIndices;
      if (FindElementAtOffset(OrigBase->getType(),
                              Offset.getSExtValue(),
                              NewIndices)) {
        // If we were able to index down into an element, create the GEP
        // and bitcast the result.  This eliminates one bitcast, potentially
        // two.
        Value *NGEP = cast<GEPOperator>(GEP)->isInBounds() ?
          Builder->CreateInBoundsGEP(OrigBase, NewIndices) :
          Builder->CreateGEP(OrigBase, NewIndices);
        NGEP->takeName(GEP);

        if (isa<BitCastInst>(CI))
          return new BitCastInst(NGEP, CI.getType());
        assert(isa<PtrToIntInst>(CI));
        return new PtrToIntInst(NGEP, CI.getType());
      }
    }
  }

  return commonCastTransforms(CI);
}

Instruction* llvm::InstCombiner::commonRemTransforms

(

BinaryOperator &

I

)

Instruction * InstCombiner::commonShiftTransforms

(

BinaryOperator &

I

)

Definition at line 24 of file InstCombineShifts.cpp.

                                                                  {
  assert(I.getOperand(1)->getType() == I.getOperand(0)->getType());
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  // See if we can fold away this shift.
  if (SimplifyDemandedInstructionBits(I))
    return &I;

  // Try to fold constant and into select arguments.
  if (isa<Constant>(Op0))
    if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;

  if (Constant *CUI = dyn_cast<Constant>(Op1))
    if (Instruction *Res = FoldShiftByConstant(Op0, CUI, I))
      return Res;

  // X shift (A srem B) -> X shift (A and B-1) iff B is a power of 2.
  // Because shifts by negative values (which could occur if A were negative)
  // are undefined.
  Value *A; const APInt *B;
  if (Op1->hasOneUse() && match(Op1, m_SRem(m_Value(A), m_Power2(B)))) {
    // FIXME: Should this get moved into SimplifyDemandedBits by saying we don't
    // demand the sign bit (and many others) here??
    Value *Rem = Builder->CreateAnd(A, ConstantInt::get(I.getType(), *B-1),
                                    Op1->getName());
    I.setOperand(1, Rem);
    return &I;
  }

  return nullptr;
}

void llvm::InstCombiner::computeKnownBits ( Value *  V, APInt &  KnownZero, APInt &  KnownOne, unsigned  Depth = 0  ) const

inline

Definition at line 318 of file InstCombine.h.

                                                  {
    return llvm::computeKnownBits(V, KnownZero, KnownOne, DL, Depth);
  }

unsigned llvm::InstCombiner::ComputeNumSignBits ( Value *  Op, unsigned  Depth = 0  ) const

inline

Definition at line 327 of file InstCombine.h.

                                                                   {
    return llvm::ComputeNumSignBits(Op, DL, Depth);
  }

Pass * Pass::createPass ( AnalysisID  ID )

staticinherited

Definition at line 195 of file Pass.cpp.

                                    {
  const PassInfo *PI = PassRegistry::getPassRegistry()->getPassInfo(ID);
  if (!PI)
    return nullptr;
  return PI->createPass();
}

Pass * FunctionPass::createPrinterPass ( raw_ostream &  O, const std::string &  Banner  ) const

overridevirtualinherited

createPrinterPass - Get a function printer pass.

Implements llvm::Pass.

Definition at line 134 of file Pass.cpp.

                                                                     {
  return createPrintFunctionPass(O, Banner);
}

Value * InstCombiner::Descale ( Value *  Val, APInt  Scale, bool &  NoSignedWrap  )

private

Descale - Return a value X such that Val = X * Scale, or null if none. If the multiplication is known not to overflow then NoSignedWrap is set.

Definition at line 940 of file InstructionCombining.cpp.

                                                                        {
  assert(isa<IntegerType>(Val->getType()) && "Can only descale integers!");
  assert(cast<IntegerType>(Val->getType())->getBitWidth() ==
         Scale.getBitWidth() && "Scale not compatible with value!");

  // If Val is zero or Scale is one then Val = Val * Scale.
  if (match(Val, m_Zero()) || Scale == 1) {
    NoSignedWrap = true;
    return Val;
  }

  // If Scale is zero then it does not divide Val.
  if (Scale.isMinValue())
    return nullptr;

  // Look through chains of multiplications, searching for a constant that is
  // divisible by Scale.  For example, descaling X*(Y*(Z*4)) by a factor of 4
  // will find the constant factor 4 and produce X*(Y*Z).  Descaling X*(Y*8) by
  // a factor of 4 will produce X*(Y*2).  The principle of operation is to bore
  // down from Val:
  //
  //     Val = M1 * X          ||   Analysis starts here and works down
  //      M1 = M2 * Y          ||   Doesn't descend into terms with more
  //      M2 =  Z * 4          \/   than one use
  //
  // Then to modify a term at the bottom:
  //
  //     Val = M1 * X
  //      M1 =  Z * Y          ||   Replaced M2 with Z
  //
  // Then to work back up correcting nsw flags.

  // Op - the term we are currently analyzing.  Starts at Val then drills down.
  // Replaced with its descaled value before exiting from the drill down loop.
  Value *Op = Val;

  // Parent - initially null, but after drilling down notes where Op came from.
  // In the example above, Parent is (Val, 0) when Op is M1, because M1 is the
  // 0'th operand of Val.
  std::pair<Instruction*, unsigned> Parent;

  // RequireNoSignedWrap - Set if the transform requires a descaling at deeper
  // levels that doesn't overflow.
  bool RequireNoSignedWrap = false;

  // logScale - log base 2 of the scale.  Negative if not a power of 2.
  int32_t logScale = Scale.exactLogBase2();

  for (;; Op = Parent.first->getOperand(Parent.second)) { // Drill down

    if (ConstantInt *CI = dyn_cast<ConstantInt>(Op)) {
      // If Op is a constant divisible by Scale then descale to the quotient.
      APInt Quotient(Scale), Remainder(Scale); // Init ensures right bitwidth.
      APInt::sdivrem(CI->getValue(), Scale, Quotient, Remainder);
      if (!Remainder.isMinValue())
        // Not divisible by Scale.
        return nullptr;
      // Replace with the quotient in the parent.
      Op = ConstantInt::get(CI->getType(), Quotient);
      NoSignedWrap = true;
      break;
    }

    if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Op)) {

      if (BO->getOpcode() == Instruction::Mul) {
        // Multiplication.
        NoSignedWrap = BO->hasNoSignedWrap();
        if (RequireNoSignedWrap && !NoSignedWrap)
          return nullptr;

        // There are three cases for multiplication: multiplication by exactly
        // the scale, multiplication by a constant different to the scale, and
        // multiplication by something else.
        Value *LHS = BO->getOperand(0);
        Value *RHS = BO->getOperand(1);

        if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
          // Multiplication by a constant.
          if (CI->getValue() == Scale) {
            // Multiplication by exactly the scale, replace the multiplication
            // by its left-hand side in the parent.
            Op = LHS;
            break;
          }

          // Otherwise drill down into the constant.
          if (!Op->hasOneUse())
            return nullptr;

          Parent = std::make_pair(BO, 1);
          continue;
        }

        // Multiplication by something else. Drill down into the left-hand side
        // since that's where the reassociate pass puts the good stuff.
        if (!Op->hasOneUse())
          return nullptr;

        Parent = std::make_pair(BO, 0);
        continue;
      }

      if (logScale > 0 && BO->getOpcode() == Instruction::Shl &&
          isa<ConstantInt>(BO->getOperand(1))) {
        // Multiplication by a power of 2.
        NoSignedWrap = BO->hasNoSignedWrap();
        if (RequireNoSignedWrap && !NoSignedWrap)
          return nullptr;

        Value *LHS = BO->getOperand(0);
        int32_t Amt = cast<ConstantInt>(BO->getOperand(1))->
          getLimitedValue(Scale.getBitWidth());
        // Op = LHS << Amt.

        if (Amt == logScale) {
          // Multiplication by exactly the scale, replace the multiplication
          // by its left-hand side in the parent.
          Op = LHS;
          break;
        }
        if (Amt < logScale || !Op->hasOneUse())
          return nullptr;

        // Multiplication by more than the scale.  Reduce the multiplying amount
        // by the scale in the parent.
        Parent = std::make_pair(BO, 1);
        Op = ConstantInt::get(BO->getType(), Amt - logScale);
        break;
      }
    }

    if (!Op->hasOneUse())
      return nullptr;

    if (CastInst *Cast = dyn_cast<CastInst>(Op)) {
      if (Cast->getOpcode() == Instruction::SExt) {
        // Op is sign-extended from a smaller type, descale in the smaller type.
        unsigned SmallSize = Cast->getSrcTy()->getPrimitiveSizeInBits();
        APInt SmallScale = Scale.trunc(SmallSize);
        // Suppose Op = sext X, and we descale X as Y * SmallScale.  We want to
        // descale Op as (sext Y) * Scale.  In order to have
        //   sext (Y * SmallScale) = (sext Y) * Scale
        // some conditions need to hold however: SmallScale must sign-extend to
        // Scale and the multiplication Y * SmallScale should not overflow.
        if (SmallScale.sext(Scale.getBitWidth()) != Scale)
          // SmallScale does not sign-extend to Scale.
          return nullptr;
        assert(SmallScale.exactLogBase2() == logScale);
        // Require that Y * SmallScale must not overflow.
        RequireNoSignedWrap = true;

        // Drill down through the cast.
        Parent = std::make_pair(Cast, 0);
        Scale = SmallScale;
        continue;
      }

      if (Cast->getOpcode() == Instruction::Trunc) {
        // Op is truncated from a larger type, descale in the larger type.
        // Suppose Op = trunc X, and we descale X as Y * sext Scale.  Then
        //   trunc (Y * sext Scale) = (trunc Y) * Scale
        // always holds.  However (trunc Y) * Scale may overflow even if
        // trunc (Y * sext Scale) does not, so nsw flags need to be cleared
        // from this point up in the expression (see later).
        if (RequireNoSignedWrap)
          return nullptr;

        // Drill down through the cast.
        unsigned LargeSize = Cast->getSrcTy()->getPrimitiveSizeInBits();
        Parent = std::make_pair(Cast, 0);
        Scale = Scale.sext(LargeSize);
        if (logScale + 1 == (int32_t)Cast->getType()->getPrimitiveSizeInBits())
          logScale = -1;
        assert(Scale.exactLogBase2() == logScale);
        continue;
      }
    }

    // Unsupported expression, bail out.
    return nullptr;
  }

  // If Op is zero then Val = Op * Scale.
  if (match(Op, m_Zero())) {
    NoSignedWrap = true;
    return Op;
  }

  // We know that we can successfully descale, so from here on we can safely
  // modify the IR.  Op holds the descaled version of the deepest term in the
  // expression.  NoSignedWrap is 'true' if multiplying Op by Scale is known
  // not to overflow.

  if (!Parent.first)
    // The expression only had one term.
    return Op;

  // Rewrite the parent using the descaled version of its operand.
  assert(Parent.first->hasOneUse() && "Drilled down when more than one use!");
  assert(Op != Parent.first->getOperand(Parent.second) &&
         "Descaling was a no-op?");
  Parent.first->setOperand(Parent.second, Op);
  Worklist.Add(Parent.first);

  // Now work back up the expression correcting nsw flags.  The logic is based
  // on the following observation: if X * Y is known not to overflow as a signed
  // multiplication, and Y is replaced by a value Z with smaller absolute value,
  // then X * Z will not overflow as a signed multiplication either.  As we work
  // our way up, having NoSignedWrap 'true' means that the descaled value at the
  // current level has strictly smaller absolute value than the original.
  Instruction *Ancestor = Parent.first;
  do {
    if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Ancestor)) {
      // If the multiplication wasn't nsw then we can't say anything about the
      // value of the descaled multiplication, and we have to clear nsw flags
      // from this point on up.
      bool OpNoSignedWrap = BO->hasNoSignedWrap();
      NoSignedWrap &= OpNoSignedWrap;
      if (NoSignedWrap != OpNoSignedWrap) {
        BO->setHasNoSignedWrap(NoSignedWrap);
        Worklist.Add(Ancestor);
      }
    } else if (Ancestor->getOpcode() == Instruction::Trunc) {
      // The fact that the descaled input to the trunc has smaller absolute
      // value than the original input doesn't tell us anything useful about
      // the absolute values of the truncations.
      NoSignedWrap = false;
    }
    assert((Ancestor->getOpcode() != Instruction::SExt || NoSignedWrap) &&
           "Failed to keep proper track of nsw flags while drilling down?");

    if (Ancestor == Val)
      // Got to the top, all done!
      return Val;

    // Move up one level in the expression.
    assert(Ancestor->hasOneUse() && "Drilled down when more than one use!");
    Ancestor = Ancestor->user_back();
  } while (1);
}

virtual bool llvm::Pass::doFinalization ( Module &  )

inlinevirtualinherited

doFinalization - Virtual method overriden by subclasses to do any necessary clean up after all passes have run.

Reimplemented in llvm::FPPassManager, llvm::legacy::FunctionPassManagerImpl, llvm::NVPTXAsmPrinter, llvm::MachineModuleInfo, llvm::AsmPrinter, legup::MinimizeBitwidth, legup::SwOnly, legup::LegupPass, and legup::LiveVariableAnalysis.

Definition at line 116 of file Pass.h.

{ return false; }

virtual bool llvm::Pass::doInitialization ( Module &  )

inlinevirtualinherited

doInitialization - Virtual method overridden by subclasses to do any necessary initialization before any pass is run.

Reimplemented in llvm::FPPassManager, llvm::legacy::FunctionPassManagerImpl, llvm::NVPTXAsmPrinter, llvm::MachineModuleInfo, llvm::AsmPrinter, legup::MinimizeBitwidth, legup::MemProf, legup::SwOnly, legup::LegupPass, llvm::MachineFunctionAnalysis, legup::LegUp, legup::LegUpTrackBB, legup::LiveVariableAnalysis, legup::LegUpWatch, and legup::PostLTO.

Definition at line 111 of file Pass.h.

{ return false; }

bool InstCombiner::DoOneIteration	(	Function &	F,
		unsigned	ItNum
	)

Definition at line 2638 of file InstructionCombining.cpp.

                                                                 {
  MadeIRChange = false;

  DEBUG(dbgs() << "\n\nINSTCOMBINE ITERATION #" << Iteration << " on "
               << F.getName() << "\n");

  {
    // Do a depth-first traversal of the function, populate the worklist with
    // the reachable instructions.  Ignore blocks that are not reachable.  Keep
    // track of which blocks we visit.
    SmallPtrSet<BasicBlock*, 64> Visited;
    MadeIRChange |= AddReachableCodeToWorklist(F.begin(), Visited, *this, DL,
                                               TLI);

    // Do a quick scan over the function.  If we find any blocks that are
    // unreachable, remove any instructions inside of them.  This prevents
    // the instcombine code from having to deal with some bad special cases.
    for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
      if (Visited.count(BB)) continue;

      // Delete the instructions backwards, as it has a reduced likelihood of
      // having to update as many def-use and use-def chains.
      Instruction *EndInst = BB->getTerminator(); // Last not to be deleted.
      while (EndInst != BB->begin()) {
        // Delete the next to last instruction.
        BasicBlock::iterator I = EndInst;
        Instruction *Inst = --I;
        if (!Inst->use_empty())
          Inst->replaceAllUsesWith(UndefValue::get(Inst->getType()));
        if (isa<LandingPadInst>(Inst)) {
          EndInst = Inst;
          continue;
        }
        if (!isa<DbgInfoIntrinsic>(Inst)) {
          ++NumDeadInst;
          MadeIRChange = true;
        }
        Inst->eraseFromParent();
      }
    }
  }

  while (!Worklist.isEmpty()) {
    Instruction *I = Worklist.RemoveOne();
    if (I == nullptr) continue;  // skip null values.

    // Check to see if we can DCE the instruction.
    if (isInstructionTriviallyDead(I, TLI)) {
      DEBUG(dbgs() << "IC: DCE: " << *I << '\n');
      EraseInstFromFunction(*I);
      ++NumDeadInst;
      MadeIRChange = true;
      continue;
    }

    // Instruction isn't dead, see if we can constant propagate it.
    if (!I->use_empty() && isa<Constant>(I->getOperand(0)))
      if (Constant *C = ConstantFoldInstruction(I, DL, TLI)) {
        DEBUG(dbgs() << "IC: ConstFold to: " << *C << " from: " << *I << '\n');

        // Add operands to the worklist.
        ReplaceInstUsesWith(*I, C);
        ++NumConstProp;
        EraseInstFromFunction(*I);
        MadeIRChange = true;
        continue;
      }

    // See if we can trivially sink this instruction to a successor basic block.
    if (I->hasOneUse()) {
      BasicBlock *BB = I->getParent();
      Instruction *UserInst = cast<Instruction>(*I->user_begin());
      BasicBlock *UserParent;

      // Get the block the use occurs in.
      if (PHINode *PN = dyn_cast<PHINode>(UserInst))
        UserParent = PN->getIncomingBlock(*I->use_begin());
      else
        UserParent = UserInst->getParent();

      if (UserParent != BB) {
        bool UserIsSuccessor = false;
        // See if the user is one of our successors.
        for (succ_iterator SI = succ_begin(BB), E = succ_end(BB); SI != E; ++SI)
          if (*SI == UserParent) {
            UserIsSuccessor = true;
            break;
          }

        // If the user is one of our immediate successors, and if that successor
        // only has us as a predecessors (we'd have to split the critical edge
        // otherwise), we can keep going.
        if (UserIsSuccessor && UserParent->getSinglePredecessor()) {
          // Okay, the CFG is simple enough, try to sink this instruction.
          if (TryToSinkInstruction(I, UserParent)) {
            MadeIRChange = true;
            // We'll add uses of the sunk instruction below, but since sinking
            // can expose opportunities for it's *operands* add them to the
            // worklist
            for (Use &U : I->operands())
              if (Instruction *OpI = dyn_cast<Instruction>(U.get()))
                Worklist.Add(OpI);
          }
        }
      }
    }

    // Now that we have an instruction, try combining it to simplify it.
    Builder->SetInsertPoint(I->getParent(), I);
    Builder->SetCurrentDebugLocation(I->getDebugLoc());

#ifndef NDEBUG
    std::string OrigI;
#endif
    DEBUG(raw_string_ostream SS(OrigI); I->print(SS); OrigI = SS.str(););
    DEBUG(dbgs() << "IC: Visiting: " << OrigI << '\n');

    if (Instruction *Result = visit(*I)) {
      ++NumCombined;
      // Should we replace the old instruction with a new one?
      if (Result != I) {
        DEBUG(dbgs() << "IC: Old = " << *I << '\n'
                     << "    New = " << *Result << '\n');

        if (!I->getDebugLoc().isUnknown())
          Result->setDebugLoc(I->getDebugLoc());
        // Everything uses the new instruction now.
        I->replaceAllUsesWith(Result);

        // Move the name to the new instruction first.
        Result->takeName(I);

        // Push the new instruction and any users onto the worklist.
        Worklist.Add(Result);
        Worklist.AddUsersToWorkList(*Result);

        // Insert the new instruction into the basic block...
        BasicBlock *InstParent = I->getParent();
        BasicBlock::iterator InsertPos = I;

        // If we replace a PHI with something that isn't a PHI, fix up the
        // insertion point.
        if (!isa<PHINode>(Result) && isa<PHINode>(InsertPos))
          InsertPos = InstParent->getFirstInsertionPt();

        InstParent->getInstList().insert(InsertPos, Result);

        EraseInstFromFunction(*I);
      } else {
#ifndef NDEBUG
        DEBUG(dbgs() << "IC: Mod = " << OrigI << '\n'
                     << "    New = " << *I << '\n');
#endif

        // If the instruction was modified, it's possible that it is now dead.
        // if so, remove it.
        if (isInstructionTriviallyDead(I, TLI)) {
          EraseInstFromFunction(*I);
        } else {
          Worklist.Add(I);
          Worklist.AddUsersToWorkList(*I);
        }
      }
      MadeIRChange = true;
    }
  }

  Worklist.Zap();
  return MadeIRChange;
}

void Pass::dump ( ) const

inherited

Definition at line 116 of file Pass.cpp.

                      {
  print(dbgs(), nullptr);
}

void Pass::dumpPassStructure ( unsigned  Offset = 0 )

virtualinherited

Reimplemented in llvm::FPPassManager, llvm::LPPassManager, and llvm::RGPassManager.

Definition at line 53 of file Pass.cpp.

                                            {
  dbgs().indent(Offset*2) << getPassName() << "\n";
}

Value * InstCombiner::dyn_castFNegVal ( Value *  V, bool  NoSignedZero = false  ) const

private

Definition at line 631 of file InstructionCombining.cpp.

                                                                        {
  if (BinaryOperator::isFNeg(V, IgnoreZeroSign))
    return BinaryOperator::getFNegArgument(V);

  // Constants can be considered to be negated values if they can be folded.
  if (ConstantFP *C = dyn_cast<ConstantFP>(V))
    return ConstantExpr::getFNeg(C);

  if (ConstantDataVector *C = dyn_cast<ConstantDataVector>(V))
    if (C->getType()->getElementType()->isFloatingPointTy())
      return ConstantExpr::getFNeg(C);

  return nullptr;
}

Value * InstCombiner::dyn_castNegVal ( Value *  V ) const

private

Definition at line 612 of file InstructionCombining.cpp.

                                                  {
  if (BinaryOperator::isNeg(V))
    return BinaryOperator::getNegArgument(V);

  // Constants can be considered to be negated values if they can be folded.
  if (ConstantInt *C = dyn_cast<ConstantInt>(V))
    return ConstantExpr::getNeg(C);

  if (ConstantDataVector *C = dyn_cast<ConstantDataVector>(V))
    if (C->getType()->getElementType()->isIntegerTy())
      return ConstantExpr::getNeg(C);

  return nullptr;
}

Value * InstCombiner::EmitGEPOffset ( User *  GEP )

private

Definition at line 98 of file InstructionCombining.cpp.

                                            {
  return llvm::EmitGEPOffset(Builder, *getDataLayout(), GEP);
}

Instruction* llvm::InstCombiner::EraseInstFromFunction ( Instruction &  I )

inline

Definition at line 301 of file InstCombine.h.

                                                     {
    DEBUG(dbgs() << "IC: ERASE " << I << '\n');

    assert(I.use_empty() && "Cannot erase instruction that is used!");
    // Make sure that we reprocess all operands now that we reduced their
    // use counts.
    if (I.getNumOperands() < 8) {
      for (User::op_iterator i = I.op_begin(), e = I.op_end(); i != e; ++i)
        if (Instruction *Op = dyn_cast<Instruction>(*i))
          Worklist.Add(Op);
    }
    Worklist.Remove(&I);
    I.eraseFromParent();
    MadeIRChange = true;
    return nullptr; // Don't do anything with FI
  }

Value * InstCombiner::EvaluateInDifferentElementOrder ( Value *  V, ArrayRef< int >  Mask  )

private

Definition at line 744 of file InstCombineVectorOps.cpp.

                                                                          {
  // Mask.size() does not need to be equal to the number of vector elements.

  assert(V->getType()->isVectorTy() && "can't reorder non-vector elements");
  if (isa<UndefValue>(V)) {
    return UndefValue::get(VectorType::get(V->getType()->getScalarType(),
                                           Mask.size()));
  }
  if (isa<ConstantAggregateZero>(V)) {
    return ConstantAggregateZero::get(
               VectorType::get(V->getType()->getScalarType(),
                               Mask.size()));
  }
  if (Constant *C = dyn_cast<Constant>(V)) {
    SmallVector<Constant *, 16> MaskValues;
    for (int i = 0, e = Mask.size(); i != e; ++i) {
      if (Mask[i] == -1)
        MaskValues.push_back(UndefValue::get(Builder->getInt32Ty()));
      else
        MaskValues.push_back(Builder->getInt32(Mask[i]));
    }
    return ConstantExpr::getShuffleVector(C, UndefValue::get(C->getType()),
                                          ConstantVector::get(MaskValues));
  }

  Instruction *I = cast<Instruction>(V);
  switch (I->getOpcode()) {
    case Instruction::Add:
    case Instruction::FAdd:
    case Instruction::Sub:
    case Instruction::FSub:
    case Instruction::Mul:
    case Instruction::FMul:
    case Instruction::UDiv:
    case Instruction::SDiv:
    case Instruction::FDiv:
    case Instruction::URem:
    case Instruction::SRem:
    case Instruction::FRem:
    case Instruction::Shl:
    case Instruction::LShr:
    case Instruction::AShr:
    case Instruction::And:
    case Instruction::Or:
    case Instruction::Xor:
    case Instruction::ICmp:
    case Instruction::FCmp:
    case Instruction::Trunc:
    case Instruction::ZExt:
    case Instruction::SExt:
    case Instruction::FPToUI:
    case Instruction::FPToSI:
    case Instruction::UIToFP:
    case Instruction::SIToFP:
    case Instruction::FPTrunc:
    case Instruction::FPExt:
    case Instruction::Select:
    case Instruction::GetElementPtr: {
      SmallVector<Value*, 8> NewOps;
      bool NeedsRebuild = (Mask.size() != I->getType()->getVectorNumElements());
      for (int i = 0, e = I->getNumOperands(); i != e; ++i) {
        Value *V = EvaluateInDifferentElementOrder(I->getOperand(i), Mask);
        NewOps.push_back(V);
        NeedsRebuild |= (V != I->getOperand(i));
      }
      if (NeedsRebuild) {
        return BuildNew(I, NewOps);
      }
      return I;
    }
    case Instruction::InsertElement: {
      int Element = cast<ConstantInt>(I->getOperand(2))->getLimitedValue();

      // The insertelement was inserting at Element. Figure out which element
      // that becomes after shuffling. The answer is guaranteed to be unique
      // by CanEvaluateShuffled.
      bool Found = false;
      int Index = 0;
      for (int e = Mask.size(); Index != e; ++Index) {
        if (Mask[Index] == Element) {
          Found = true;
          break;
        }
      }

      // If element is not in Mask, no need to handle the operand 1 (element to
      // be inserted). Just evaluate values in operand 0 according to Mask.
      if (!Found)
        return EvaluateInDifferentElementOrder(I->getOperand(0), Mask);

      Value *V = EvaluateInDifferentElementOrder(I->getOperand(0), Mask);
      return InsertElementInst::Create(V, I->getOperand(1),
                                       Builder->getInt32(Index), "", I);
    }
  }
  llvm_unreachable("failed to reorder elements of vector instruction!");
}

Value * InstCombiner::EvaluateInDifferentType ( Value *  V, Type *  Ty, bool  isSigned  )

private

EvaluateInDifferentType - Given an expression that CanEvaluateTruncated or CanEvaluateSExtd returns true for, actually insert the code to evaluate the expression.

Definition at line 163 of file InstCombineCasts.cpp.

                                                            {
  if (Constant *C = dyn_cast<Constant>(V)) {
    C = ConstantExpr::getIntegerCast(C, Ty, isSigned /*Sext or ZExt*/);
    // If we got a constantexpr back, try to simplify it with DL info.
    if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
      C = ConstantFoldConstantExpression(CE, DL, TLI);
    return C;
  }

  // Otherwise, it must be an instruction.
  Instruction *I = cast<Instruction>(V);
  Instruction *Res = nullptr;
  unsigned Opc = I->getOpcode();
  switch (Opc) {
  case Instruction::Add:
  case Instruction::Sub:
  case Instruction::Mul:
  case Instruction::And:
  case Instruction::Or:
  case Instruction::Xor:
  case Instruction::AShr:
  case Instruction::LShr:
  case Instruction::Shl:
  case Instruction::UDiv:
  case Instruction::URem: {
    Value *LHS = EvaluateInDifferentType(I->getOperand(0), Ty, isSigned);
    Value *RHS = EvaluateInDifferentType(I->getOperand(1), Ty, isSigned);
    Res = BinaryOperator::Create((Instruction::BinaryOps)Opc, LHS, RHS);
    break;
  }
  case Instruction::Trunc:
  case Instruction::ZExt:
  case Instruction::SExt:
    // If the source type of the cast is the type we're trying for then we can
    // just return the source.  There's no need to insert it because it is not
    // new.
    if (I->getOperand(0)->getType() == Ty)
      return I->getOperand(0);

    // Otherwise, must be the same type of cast, so just reinsert a new one.
    // This also handles the case of zext(trunc(x)) -> zext(x).
    Res = CastInst::CreateIntegerCast(I->getOperand(0), Ty,
                                      Opc == Instruction::SExt);
    break;
  case Instruction::Select: {
    Value *True = EvaluateInDifferentType(I->getOperand(1), Ty, isSigned);
    Value *False = EvaluateInDifferentType(I->getOperand(2), Ty, isSigned);
    Res = SelectInst::Create(I->getOperand(0), True, False);
    break;
  }
  case Instruction::PHI: {
    PHINode *OPN = cast<PHINode>(I);
    PHINode *NPN = PHINode::Create(Ty, OPN->getNumIncomingValues());
    for (unsigned i = 0, e = OPN->getNumIncomingValues(); i != e; ++i) {
      Value *V =EvaluateInDifferentType(OPN->getIncomingValue(i), Ty, isSigned);
      NPN->addIncoming(V, OPN->getIncomingBlock(i));
    }
    Res = NPN;
    break;
  }
  default:
    // TODO: Can handle more cases here.
    llvm_unreachable("Unreachable!");
  }

  Res->takeName(I);
  return InsertNewInstWith(Res, *I);
}

Type * InstCombiner::FindElementAtOffset ( Type *  PtrTy, int64_t  Offset, SmallVectorImpl< Value * > &  NewIndices  )

private

FindElementAtOffset - Given a pointer type and a constant offset, determine whether or not there is a sequence of GEP indices into the pointed type that will land us at the specified offset. If so, fill them into NewIndices and return the resultant element type, otherwise return null.

Definition at line 865 of file InstructionCombining.cpp.

                                                                             {
  assert(PtrTy->isPtrOrPtrVectorTy());

  if (!DL)
    return nullptr;

  Type *Ty = PtrTy->getPointerElementType();
  if (!Ty->isSized())
    return nullptr;

  // Start with the index over the outer type.  Note that the type size
  // might be zero (even if the offset isn't zero) if the indexed type
  // is something like [0 x {int, int}]
  Type *IntPtrTy = DL->getIntPtrType(PtrTy);
  int64_t FirstIdx = 0;
  if (int64_t TySize = DL->getTypeAllocSize(Ty)) {
    FirstIdx = Offset/TySize;
    Offset -= FirstIdx*TySize;

    // Handle hosts where % returns negative instead of values [0..TySize).
    if (Offset < 0) {
      --FirstIdx;
      Offset += TySize;
      assert(Offset >= 0);
    }
    assert((uint64_t)Offset < (uint64_t)TySize && "Out of range offset");
  }

  NewIndices.push_back(ConstantInt::get(IntPtrTy, FirstIdx));

  // Index into the types.  If we fail, set OrigBase to null.
  while (Offset) {
    // Indexing into tail padding between struct/array elements.
    if (uint64_t(Offset*8) >= DL->getTypeSizeInBits(Ty))
      return nullptr;

    if (StructType *STy = dyn_cast<StructType>(Ty)) {
      const StructLayout *SL = DL->getStructLayout(STy);
      assert(Offset < (int64_t)SL->getSizeInBytes() &&
             "Offset must stay within the indexed type");

      unsigned Elt = SL->getElementContainingOffset(Offset);
      NewIndices.push_back(ConstantInt::get(Type::getInt32Ty(Ty->getContext()),
                                            Elt));

      Offset -= SL->getElementOffset(Elt);
      Ty = STy->getElementType(Elt);
    } else if (ArrayType *AT = dyn_cast<ArrayType>(Ty)) {
      uint64_t EltSize = DL->getTypeAllocSize(AT->getElementType());
      assert(EltSize && "Cannot index into a zero-sized array");
      NewIndices.push_back(ConstantInt::get(IntPtrTy,Offset/EltSize));
      Offset %= EltSize;
      Ty = AT->getElementType();
    } else {
      // Otherwise, we can't index into the middle of this atomic type, bail.
      return nullptr;
    }
  }

  return Ty;
}

Value * InstCombiner::FoldAndOfFCmps	(	FCmpInst *	LHS,
		FCmpInst *	RHS
	)

FoldAndOfFCmps - Optimize (fcmp)&(fcmp). NOTE: Unlike the rest of instcombine, this returns a Value which should already be inserted into the function.

Definition at line 1027 of file InstCombineAndOrXor.cpp.

                                                                {
  if (LHS->getPredicate() == FCmpInst::FCMP_ORD &&
      RHS->getPredicate() == FCmpInst::FCMP_ORD) {
    if (LHS->getOperand(0)->getType() != RHS->getOperand(0)->getType())
      return nullptr;

    // (fcmp ord x, c) & (fcmp ord y, c)  -> (fcmp ord x, y)
    if (ConstantFP *LHSC = dyn_cast<ConstantFP>(LHS->getOperand(1)))
      if (ConstantFP *RHSC = dyn_cast<ConstantFP>(RHS->getOperand(1))) {
        // If either of the constants are nans, then the whole thing returns
        // false.
        if (LHSC->getValueAPF().isNaN() || RHSC->getValueAPF().isNaN())
          return Builder->getFalse();
        return Builder->CreateFCmpORD(LHS->getOperand(0), RHS->getOperand(0));
      }

    // Handle vector zeros.  This occurs because the canonical form of
    // "fcmp ord x,x" is "fcmp ord x, 0".
    if (isa<ConstantAggregateZero>(LHS->getOperand(1)) &&
        isa<ConstantAggregateZero>(RHS->getOperand(1)))
      return Builder->CreateFCmpORD(LHS->getOperand(0), RHS->getOperand(0));
    return nullptr;
  }

  Value *Op0LHS = LHS->getOperand(0), *Op0RHS = LHS->getOperand(1);
  Value *Op1LHS = RHS->getOperand(0), *Op1RHS = RHS->getOperand(1);
  FCmpInst::Predicate Op0CC = LHS->getPredicate(), Op1CC = RHS->getPredicate();


  if (Op0LHS == Op1RHS && Op0RHS == Op1LHS) {
    // Swap RHS operands to match LHS.
    Op1CC = FCmpInst::getSwappedPredicate(Op1CC);
    std::swap(Op1LHS, Op1RHS);
  }

  if (Op0LHS == Op1LHS && Op0RHS == Op1RHS) {
    // Simplify (fcmp cc0 x, y) & (fcmp cc1 x, y).
    if (Op0CC == Op1CC)
      return Builder->CreateFCmp((FCmpInst::Predicate)Op0CC, Op0LHS, Op0RHS);
    if (Op0CC == FCmpInst::FCMP_FALSE || Op1CC == FCmpInst::FCMP_FALSE)
      return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
    if (Op0CC == FCmpInst::FCMP_TRUE)
      return RHS;
    if (Op1CC == FCmpInst::FCMP_TRUE)
      return LHS;

    bool Op0Ordered;
    bool Op1Ordered;
    unsigned Op0Pred = getFCmpCode(Op0CC, Op0Ordered);
    unsigned Op1Pred = getFCmpCode(Op1CC, Op1Ordered);
    // uno && ord -> false
    if (Op0Pred == 0 && Op1Pred == 0 && Op0Ordered != Op1Ordered)
        return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
    if (Op1Pred == 0) {
      std::swap(LHS, RHS);
      std::swap(Op0Pred, Op1Pred);
      std::swap(Op0Ordered, Op1Ordered);
    }
    if (Op0Pred == 0) {
      // uno && ueq -> uno && (uno || eq) -> uno
      // ord && olt -> ord && (ord && lt) -> olt
      if (!Op0Ordered && (Op0Ordered == Op1Ordered))
        return LHS;
      if (Op0Ordered && (Op0Ordered == Op1Ordered))
        return RHS;

      // uno && oeq -> uno && (ord && eq) -> false
      if (!Op0Ordered)
        return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
      // ord && ueq -> ord && (uno || eq) -> oeq
      return getFCmpValue(true, Op1Pred, Op0LHS, Op0RHS, Builder);
    }
  }

  return nullptr;
}

Value * InstCombiner::FoldAndOfICmps	(	ICmpInst *	LHS,
		ICmpInst *	RHS
	)

FoldAndOfICmps - Fold (icmp)&(icmp) if possible.

Definition at line 791 of file InstCombineAndOrXor.cpp.

                                                                {
  ICmpInst::Predicate LHSCC = LHS->getPredicate(), RHSCC = RHS->getPredicate();

  // (icmp1 A, B) & (icmp2 A, B) --> (icmp3 A, B)
  if (PredicatesFoldable(LHSCC, RHSCC)) {
    if (LHS->getOperand(0) == RHS->getOperand(1) &&
        LHS->getOperand(1) == RHS->getOperand(0))
      LHS->swapOperands();
    if (LHS->getOperand(0) == RHS->getOperand(0) &&
        LHS->getOperand(1) == RHS->getOperand(1)) {
      Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
      unsigned Code = getICmpCode(LHS) & getICmpCode(RHS);
      bool isSigned = LHS->isSigned() || RHS->isSigned();
      return getNewICmpValue(isSigned, Code, Op0, Op1, Builder);
    }
  }

  // handle (roughly):  (icmp eq (A & B), C) & (icmp eq (A & D), E)
  if (Value *V = foldLogOpOfMaskedICmps(LHS, RHS, true, Builder))
    return V;

  // This only handles icmp of constants: (icmp1 A, C1) & (icmp2 B, C2).
  Value *Val = LHS->getOperand(0), *Val2 = RHS->getOperand(0);
  ConstantInt *LHSCst = dyn_cast<ConstantInt>(LHS->getOperand(1));
  ConstantInt *RHSCst = dyn_cast<ConstantInt>(RHS->getOperand(1));
  if (!LHSCst || !RHSCst) return nullptr;

  if (LHSCst == RHSCst && LHSCC == RHSCC) {
    // (icmp ult A, C) & (icmp ult B, C) --> (icmp ult (A|B), C)
    // where C is a power of 2
    if (LHSCC == ICmpInst::ICMP_ULT &&
        LHSCst->getValue().isPowerOf2()) {
      Value *NewOr = Builder->CreateOr(Val, Val2);
      return Builder->CreateICmp(LHSCC, NewOr, LHSCst);
    }

    // (icmp eq A, 0) & (icmp eq B, 0) --> (icmp eq (A|B), 0)
    if (LHSCC == ICmpInst::ICMP_EQ && LHSCst->isZero()) {
      Value *NewOr = Builder->CreateOr(Val, Val2);
      return Builder->CreateICmp(LHSCC, NewOr, LHSCst);
    }
  }

  // (trunc x) == C1 & (and x, CA) == C2 -> (and x, CA|CMAX) == C1|C2
  // where CMAX is the all ones value for the truncated type,
  // iff the lower bits of C2 and CA are zero.
  if (LHSCC == ICmpInst::ICMP_EQ && LHSCC == RHSCC &&
      LHS->hasOneUse() && RHS->hasOneUse()) {
    Value *V;
    ConstantInt *AndCst, *SmallCst = nullptr, *BigCst = nullptr;

    // (trunc x) == C1 & (and x, CA) == C2
    // (and x, CA) == C2 & (trunc x) == C1
    if (match(Val2, m_Trunc(m_Value(V))) &&
        match(Val, m_And(m_Specific(V), m_ConstantInt(AndCst)))) {
      SmallCst = RHSCst;
      BigCst = LHSCst;
    } else if (match(Val, m_Trunc(m_Value(V))) &&
               match(Val2, m_And(m_Specific(V), m_ConstantInt(AndCst)))) {
      SmallCst = LHSCst;
      BigCst = RHSCst;
    }

    if (SmallCst && BigCst) {
      unsigned BigBitSize = BigCst->getType()->getBitWidth();
      unsigned SmallBitSize = SmallCst->getType()->getBitWidth();

      // Check that the low bits are zero.
      APInt Low = APInt::getLowBitsSet(BigBitSize, SmallBitSize);
      if ((Low & AndCst->getValue()) == 0 && (Low & BigCst->getValue()) == 0) {
        Value *NewAnd = Builder->CreateAnd(V, Low | AndCst->getValue());
        APInt N = SmallCst->getValue().zext(BigBitSize) | BigCst->getValue();
        Value *NewVal = ConstantInt::get(AndCst->getType()->getContext(), N);
        return Builder->CreateICmp(LHSCC, NewAnd, NewVal);
      }
    }
  }

  // From here on, we only handle:
  //    (icmp1 A, C1) & (icmp2 A, C2) --> something simpler.
  if (Val != Val2) return nullptr;

  // ICMP_[US][GL]E X, CST is folded to ICMP_[US][GL]T elsewhere.
  if (LHSCC == ICmpInst::ICMP_UGE || LHSCC == ICmpInst::ICMP_ULE ||
      RHSCC == ICmpInst::ICMP_UGE || RHSCC == ICmpInst::ICMP_ULE ||
      LHSCC == ICmpInst::ICMP_SGE || LHSCC == ICmpInst::ICMP_SLE ||
      RHSCC == ICmpInst::ICMP_SGE || RHSCC == ICmpInst::ICMP_SLE)
    return nullptr;

  // Make a constant range that's the intersection of the two icmp ranges.
  // If the intersection is empty, we know that the result is false.
  ConstantRange LHSRange =
    ConstantRange::makeICmpRegion(LHSCC, LHSCst->getValue());
  ConstantRange RHSRange =
    ConstantRange::makeICmpRegion(RHSCC, RHSCst->getValue());

  if (LHSRange.intersectWith(RHSRange).isEmptySet())
    return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);

  // We can't fold (ugt x, C) & (sgt x, C2).
  if (!PredicatesFoldable(LHSCC, RHSCC))
    return nullptr;

  // Ensure that the larger constant is on the RHS.
  bool ShouldSwap;
  if (CmpInst::isSigned(LHSCC) ||
      (ICmpInst::isEquality(LHSCC) &&
       CmpInst::isSigned(RHSCC)))
    ShouldSwap = LHSCst->getValue().sgt(RHSCst->getValue());
  else
    ShouldSwap = LHSCst->getValue().ugt(RHSCst->getValue());

  if (ShouldSwap) {
    std::swap(LHS, RHS);
    std::swap(LHSCst, RHSCst);
    std::swap(LHSCC, RHSCC);
  }

  // At this point, we know we have two icmp instructions
  // comparing a value against two constants and and'ing the result
  // together.  Because of the above check, we know that we only have
  // icmp eq, icmp ne, icmp [su]lt, and icmp [SU]gt here. We also know
  // (from the icmp folding check above), that the two constants
  // are not equal and that the larger constant is on the RHS
  assert(LHSCst != RHSCst && "Compares not folded above?");

  switch (LHSCC) {
  default: llvm_unreachable("Unknown integer condition code!");
  case ICmpInst::ICMP_EQ:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_NE:         // (X == 13 & X != 15) -> X == 13
    case ICmpInst::ICMP_ULT:        // (X == 13 & X <  15) -> X == 13
    case ICmpInst::ICMP_SLT:        // (X == 13 & X <  15) -> X == 13
      return LHS;
    }
  case ICmpInst::ICMP_NE:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_ULT:
      if (LHSCst == SubOne(RHSCst)) // (X != 13 & X u< 14) -> X < 13
        return Builder->CreateICmpULT(Val, LHSCst);
      break;                        // (X != 13 & X u< 15) -> no change
    case ICmpInst::ICMP_SLT:
      if (LHSCst == SubOne(RHSCst)) // (X != 13 & X s< 14) -> X < 13
        return Builder->CreateICmpSLT(Val, LHSCst);
      break;                        // (X != 13 & X s< 15) -> no change
    case ICmpInst::ICMP_EQ:         // (X != 13 & X == 15) -> X == 15
    case ICmpInst::ICMP_UGT:        // (X != 13 & X u> 15) -> X u> 15
    case ICmpInst::ICMP_SGT:        // (X != 13 & X s> 15) -> X s> 15
      return RHS;
    case ICmpInst::ICMP_NE:
      // Special case to get the ordering right when the values wrap around
      // zero.
      if (LHSCst->getValue() == 0 && RHSCst->getValue().isAllOnesValue())
        std::swap(LHSCst, RHSCst);
      if (LHSCst == SubOne(RHSCst)){// (X != 13 & X != 14) -> X-13 >u 1
        Constant *AddCST = ConstantExpr::getNeg(LHSCst);
        Value *Add = Builder->CreateAdd(Val, AddCST, Val->getName()+".off");
        return Builder->CreateICmpUGT(Add, ConstantInt::get(Add->getType(), 1),
                                      Val->getName()+".cmp");
      }
      break;                        // (X != 13 & X != 15) -> no change
    }
    break;
  case ICmpInst::ICMP_ULT:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:         // (X u< 13 & X == 15) -> false
    case ICmpInst::ICMP_UGT:        // (X u< 13 & X u> 15) -> false
      return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
    case ICmpInst::ICMP_SGT:        // (X u< 13 & X s> 15) -> no change
      break;
    case ICmpInst::ICMP_NE:         // (X u< 13 & X != 15) -> X u< 13
    case ICmpInst::ICMP_ULT:        // (X u< 13 & X u< 15) -> X u< 13
      return LHS;
    case ICmpInst::ICMP_SLT:        // (X u< 13 & X s< 15) -> no change
      break;
    }
    break;
  case ICmpInst::ICMP_SLT:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_UGT:        // (X s< 13 & X u> 15) -> no change
      break;
    case ICmpInst::ICMP_NE:         // (X s< 13 & X != 15) -> X < 13
    case ICmpInst::ICMP_SLT:        // (X s< 13 & X s< 15) -> X < 13
      return LHS;
    case ICmpInst::ICMP_ULT:        // (X s< 13 & X u< 15) -> no change
      break;
    }
    break;
  case ICmpInst::ICMP_UGT:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:         // (X u> 13 & X == 15) -> X == 15
    case ICmpInst::ICMP_UGT:        // (X u> 13 & X u> 15) -> X u> 15
      return RHS;
    case ICmpInst::ICMP_SGT:        // (X u> 13 & X s> 15) -> no change
      break;
    case ICmpInst::ICMP_NE:
      if (RHSCst == AddOne(LHSCst)) // (X u> 13 & X != 14) -> X u> 14
        return Builder->CreateICmp(LHSCC, Val, RHSCst);
      break;                        // (X u> 13 & X != 15) -> no change
    case ICmpInst::ICMP_ULT:        // (X u> 13 & X u< 15) -> (X-14) <u 1
      return InsertRangeTest(Val, AddOne(LHSCst), RHSCst, false, true);
    case ICmpInst::ICMP_SLT:        // (X u> 13 & X s< 15) -> no change
      break;
    }
    break;
  case ICmpInst::ICMP_SGT:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:         // (X s> 13 & X == 15) -> X == 15
    case ICmpInst::ICMP_SGT:        // (X s> 13 & X s> 15) -> X s> 15
      return RHS;
    case ICmpInst::ICMP_UGT:        // (X s> 13 & X u> 15) -> no change
      break;
    case ICmpInst::ICMP_NE:
      if (RHSCst == AddOne(LHSCst)) // (X s> 13 & X != 14) -> X s> 14
        return Builder->CreateICmp(LHSCC, Val, RHSCst);
      break;                        // (X s> 13 & X != 15) -> no change
    case ICmpInst::ICMP_SLT:        // (X s> 13 & X s< 15) -> (X-14) s< 1
      return InsertRangeTest(Val, AddOne(LHSCst), RHSCst, true, true);
    case ICmpInst::ICMP_ULT:        // (X s> 13 & X u< 15) -> no change
      break;
    }
    break;
  }

  return nullptr;
}

Instruction * InstCombiner::FoldCmpLoadFromIndexedGlobal	(	GetElementPtrInst *	GEP,
		GlobalVariable *	GV,
		CmpInst &	ICI,
		ConstantInt *	AndCst = nullptr
	)

FoldCmpLoadFromIndexedGlobal - Called we see this pattern: cmp pred (load (gep GV, ...)), cmpcst where GV is a global variable with a constant initializer. Try to simplify this into some simple computation that does not need the load. For example we can optimize "icmp eq (load (gep "foo", 0, i)), 0" into "icmp eq i, 3".

If AndCst is non-null, then the loaded value is masked with that constant before doing the comparison. This handles cases like "A[i]&4 == 0".

TrueRangeEnd/FalseRangeEnd - In conjunction with First*Element, these define a state machine that triggers for ranges of values that the index is true or false for. This triggers on things like "abbbbc"[i] == 'b'. This is -2 when undefined, -3 when overdefined, and otherwise the last index in the range (inclusive). We use -2 for undefined here because we use relative comparisons and don't want 0-1 to match -1.

Definition at line 220 of file InstCombineCompares.cpp.

                                                                {
  // We need TD information to know the pointer size unless this is inbounds.
  if (!GEP->isInBounds() && !DL)
    return nullptr;

  Constant *Init = GV->getInitializer();
  if (!isa<ConstantArray>(Init) && !isa<ConstantDataArray>(Init))
    return nullptr;

  uint64_t ArrayElementCount = Init->getType()->getArrayNumElements();
  if (ArrayElementCount > 1024) return nullptr; // Don't blow up on huge arrays.

  // There are many forms of this optimization we can handle, for now, just do
  // the simple index into a single-dimensional array.
  //
  // Require: GEP GV, 0, i {{, constant indices}}
  if (GEP->getNumOperands() < 3 ||
      !isa<ConstantInt>(GEP->getOperand(1)) ||
      !cast<ConstantInt>(GEP->getOperand(1))->isZero() ||
      isa<Constant>(GEP->getOperand(2)))
    return nullptr;

  // Check that indices after the variable are constants and in-range for the
  // type they index.  Collect the indices.  This is typically for arrays of
  // structs.
  SmallVector<unsigned, 4> LaterIndices;

  Type *EltTy = Init->getType()->getArrayElementType();
  for (unsigned i = 3, e = GEP->getNumOperands(); i != e; ++i) {
    ConstantInt *Idx = dyn_cast<ConstantInt>(GEP->getOperand(i));
    if (!Idx) return nullptr;  // Variable index.

    uint64_t IdxVal = Idx->getZExtValue();
    if ((unsigned)IdxVal != IdxVal) return nullptr; // Too large array index.

    if (StructType *STy = dyn_cast<StructType>(EltTy))
      EltTy = STy->getElementType(IdxVal);
    else if (ArrayType *ATy = dyn_cast<ArrayType>(EltTy)) {
      if (IdxVal >= ATy->getNumElements()) return nullptr;
      EltTy = ATy->getElementType();
    } else {
      return nullptr; // Unknown type.
    }

    LaterIndices.push_back(IdxVal);
  }

  enum { Overdefined = -3, Undefined = -2 };

  // Variables for our state machines.

  // FirstTrueElement/SecondTrueElement - Used to emit a comparison of the form
  // "i == 47 | i == 87", where 47 is the first index the condition is true for,
  // and 87 is the second (and last) index.  FirstTrueElement is -2 when
  // undefined, otherwise set to the first true element.  SecondTrueElement is
  // -2 when undefined, -3 when overdefined and >= 0 when that index is true.
  int FirstTrueElement = Undefined, SecondTrueElement = Undefined;

  // FirstFalseElement/SecondFalseElement - Used to emit a comparison of the
  // form "i != 47 & i != 87".  Same state transitions as for true elements.
  int FirstFalseElement = Undefined, SecondFalseElement = Undefined;

  /// TrueRangeEnd/FalseRangeEnd - In conjunction with First*Element, these
  /// define a state machine that triggers for ranges of values that the index
  /// is true or false for.  This triggers on things like "abbbbc"[i] == 'b'.
  /// This is -2 when undefined, -3 when overdefined, and otherwise the last
  /// index in the range (inclusive).  We use -2 for undefined here because we
  /// use relative comparisons and don't want 0-1 to match -1.
  int TrueRangeEnd = Undefined, FalseRangeEnd = Undefined;

  // MagicBitvector - This is a magic bitvector where we set a bit if the
  // comparison is true for element 'i'.  If there are 64 elements or less in
  // the array, this will fully represent all the comparison results.
  uint64_t MagicBitvector = 0;


  // Scan the array and see if one of our patterns matches.
  Constant *CompareRHS = cast<Constant>(ICI.getOperand(1));
  for (unsigned i = 0, e = ArrayElementCount; i != e; ++i) {
    Constant *Elt = Init->getAggregateElement(i);
    if (!Elt) return nullptr;

    // If this is indexing an array of structures, get the structure element.
    if (!LaterIndices.empty())
      Elt = ConstantExpr::getExtractValue(Elt, LaterIndices);

    // If the element is masked, handle it.
    if (AndCst) Elt = ConstantExpr::getAnd(Elt, AndCst);

    // Find out if the comparison would be true or false for the i'th element.
    Constant *C = ConstantFoldCompareInstOperands(ICI.getPredicate(), Elt,
                                                  CompareRHS, DL, TLI);
    // If the result is undef for this element, ignore it.
    if (isa<UndefValue>(C)) {
      // Extend range state machines to cover this element in case there is an
      // undef in the middle of the range.
      if (TrueRangeEnd == (int)i-1)
        TrueRangeEnd = i;
      if (FalseRangeEnd == (int)i-1)
        FalseRangeEnd = i;
      continue;
    }

    // If we can't compute the result for any of the elements, we have to give
    // up evaluating the entire conditional.
    if (!isa<ConstantInt>(C)) return nullptr;

    // Otherwise, we know if the comparison is true or false for this element,
    // update our state machines.
    bool IsTrueForElt = !cast<ConstantInt>(C)->isZero();

    // State machine for single/double/range index comparison.
    if (IsTrueForElt) {
      // Update the TrueElement state machine.
      if (FirstTrueElement == Undefined)
        FirstTrueElement = TrueRangeEnd = i;  // First true element.
      else {
        // Update double-compare state machine.
        if (SecondTrueElement == Undefined)
          SecondTrueElement = i;
        else
          SecondTrueElement = Overdefined;

        // Update range state machine.
        if (TrueRangeEnd == (int)i-1)
          TrueRangeEnd = i;
        else
          TrueRangeEnd = Overdefined;
      }
    } else {
      // Update the FalseElement state machine.
      if (FirstFalseElement == Undefined)
        FirstFalseElement = FalseRangeEnd = i; // First false element.
      else {
        // Update double-compare state machine.
        if (SecondFalseElement == Undefined)
          SecondFalseElement = i;
        else
          SecondFalseElement = Overdefined;

        // Update range state machine.
        if (FalseRangeEnd == (int)i-1)
          FalseRangeEnd = i;
        else
          FalseRangeEnd = Overdefined;
      }
    }


    // If this element is in range, update our magic bitvector.
    if (i < 64 && IsTrueForElt)
      MagicBitvector |= 1ULL << i;

    // If all of our states become overdefined, bail out early.  Since the
    // predicate is expensive, only check it every 8 elements.  This is only
    // really useful for really huge arrays.
    if ((i & 8) == 0 && i >= 64 && SecondTrueElement == Overdefined &&
        SecondFalseElement == Overdefined && TrueRangeEnd == Overdefined &&
        FalseRangeEnd == Overdefined)
      return nullptr;
  }

  // Now that we've scanned the entire array, emit our new comparison(s).  We
  // order the state machines in complexity of the generated code.
  Value *Idx = GEP->getOperand(2);

  // If the index is larger than the pointer size of the target, truncate the
  // index down like the GEP would do implicitly.  We don't have to do this for
  // an inbounds GEP because the index can't be out of range.
  if (!GEP->isInBounds()) {
    Type *IntPtrTy = DL->getIntPtrType(GEP->getType());
    unsigned PtrSize = IntPtrTy->getIntegerBitWidth();
    if (Idx->getType()->getPrimitiveSizeInBits() > PtrSize)
      Idx = Builder->CreateTrunc(Idx, IntPtrTy);
  }

  // If the comparison is only true for one or two elements, emit direct
  // comparisons.
  if (SecondTrueElement != Overdefined) {
    // None true -> false.
    if (FirstTrueElement == Undefined)
      return ReplaceInstUsesWith(ICI, Builder->getFalse());

    Value *FirstTrueIdx = ConstantInt::get(Idx->getType(), FirstTrueElement);

    // True for one element -> 'i == 47'.
    if (SecondTrueElement == Undefined)
      return new ICmpInst(ICmpInst::ICMP_EQ, Idx, FirstTrueIdx);

    // True for two elements -> 'i == 47 | i == 72'.
    Value *C1 = Builder->CreateICmpEQ(Idx, FirstTrueIdx);
    Value *SecondTrueIdx = ConstantInt::get(Idx->getType(), SecondTrueElement);
    Value *C2 = Builder->CreateICmpEQ(Idx, SecondTrueIdx);
    return BinaryOperator::CreateOr(C1, C2);
  }

  // If the comparison is only false for one or two elements, emit direct
  // comparisons.
  if (SecondFalseElement != Overdefined) {
    // None false -> true.
    if (FirstFalseElement == Undefined)
      return ReplaceInstUsesWith(ICI, Builder->getTrue());

    Value *FirstFalseIdx = ConstantInt::get(Idx->getType(), FirstFalseElement);

    // False for one element -> 'i != 47'.
    if (SecondFalseElement == Undefined)
      return new ICmpInst(ICmpInst::ICMP_NE, Idx, FirstFalseIdx);

    // False for two elements -> 'i != 47 & i != 72'.
    Value *C1 = Builder->CreateICmpNE(Idx, FirstFalseIdx);
    Value *SecondFalseIdx = ConstantInt::get(Idx->getType(),SecondFalseElement);
    Value *C2 = Builder->CreateICmpNE(Idx, SecondFalseIdx);
    return BinaryOperator::CreateAnd(C1, C2);
  }

  // If the comparison can be replaced with a range comparison for the elements
  // where it is true, emit the range check.
  if (TrueRangeEnd != Overdefined) {
    assert(TrueRangeEnd != FirstTrueElement && "Should emit single compare");

    // Generate (i-FirstTrue) <u (TrueRangeEnd-FirstTrue+1).
    if (FirstTrueElement) {
      Value *Offs = ConstantInt::get(Idx->getType(), -FirstTrueElement);
      Idx = Builder->CreateAdd(Idx, Offs);
    }

    Value *End = ConstantInt::get(Idx->getType(),
                                  TrueRangeEnd-FirstTrueElement+1);
    return new ICmpInst(ICmpInst::ICMP_ULT, Idx, End);
  }

  // False range check.
  if (FalseRangeEnd != Overdefined) {
    assert(FalseRangeEnd != FirstFalseElement && "Should emit single compare");
    // Generate (i-FirstFalse) >u (FalseRangeEnd-FirstFalse).
    if (FirstFalseElement) {
      Value *Offs = ConstantInt::get(Idx->getType(), -FirstFalseElement);
      Idx = Builder->CreateAdd(Idx, Offs);
    }

    Value *End = ConstantInt::get(Idx->getType(),
                                  FalseRangeEnd-FirstFalseElement);
    return new ICmpInst(ICmpInst::ICMP_UGT, Idx, End);
  }


  // If a magic bitvector captures the entire comparison state
  // of this load, replace it with computation that does:
  //   ((magic_cst >> i) & 1) != 0
  {
    Type *Ty = nullptr;

    // Look for an appropriate type:
    // - The type of Idx if the magic fits
    // - The smallest fitting legal type if we have a DataLayout
    // - Default to i32
    if (ArrayElementCount <= Idx->getType()->getIntegerBitWidth())
      Ty = Idx->getType();
    else if (DL)
      Ty = DL->getSmallestLegalIntType(Init->getContext(), ArrayElementCount);
    else if (ArrayElementCount <= 32)
      Ty = Type::getInt32Ty(Init->getContext());

    if (Ty) {
      Value *V = Builder->CreateIntCast(Idx, Ty, false);
      V = Builder->CreateLShr(ConstantInt::get(Ty, MagicBitvector), V);
      V = Builder->CreateAnd(ConstantInt::get(Ty, 1), V);
      return new ICmpInst(ICmpInst::ICMP_NE, V, ConstantInt::get(Ty, 0));
    }
  }

  return nullptr;
}

Instruction * InstCombiner::FoldFCmp_IntToFP_Cst	(	FCmpInst &	I,
		Instruction *	LHSI,
		Constant *	RHSC
	)

FoldFCmp_IntToFP_Cst - Fold fcmp ([us]itofp x, cst) if possible.

Definition at line 3291 of file InstCombineCompares.cpp.

                                                                {
  if (!isa<ConstantFP>(RHSC)) return nullptr;
  const APFloat &RHS = cast<ConstantFP>(RHSC)->getValueAPF();

  // Get the width of the mantissa.  We don't want to hack on conversions that
  // might lose information from the integer, e.g. "i64 -> float"
  int MantissaWidth = LHSI->getType()->getFPMantissaWidth();
  if (MantissaWidth == -1) return nullptr;  // Unknown.

  // Check to see that the input is converted from an integer type that is small
  // enough that preserves all bits.  TODO: check here for "known" sign bits.
  // This would allow us to handle (fptosi (x >>s 62) to float) if x is i64 f.e.
  unsigned InputSize = LHSI->getOperand(0)->getType()->getScalarSizeInBits();

  // If this is a uitofp instruction, we need an extra bit to hold the sign.
  bool LHSUnsigned = isa<UIToFPInst>(LHSI);
  if (LHSUnsigned)
    ++InputSize;

  // If the conversion would lose info, don't hack on this.
  if ((int)InputSize > MantissaWidth)
    return nullptr;

  // Otherwise, we can potentially simplify the comparison.  We know that it
  // will always come through as an integer value and we know the constant is
  // not a NAN (it would have been previously simplified).
  assert(!RHS.isNaN() && "NaN comparison not already folded!");

  ICmpInst::Predicate Pred;
  switch (I.getPredicate()) {
  default: llvm_unreachable("Unexpected predicate!");
  case FCmpInst::FCMP_UEQ:
  case FCmpInst::FCMP_OEQ:
    Pred = ICmpInst::ICMP_EQ;
    break;
  case FCmpInst::FCMP_UGT:
  case FCmpInst::FCMP_OGT:
    Pred = LHSUnsigned ? ICmpInst::ICMP_UGT : ICmpInst::ICMP_SGT;
    break;
  case FCmpInst::FCMP_UGE:
  case FCmpInst::FCMP_OGE:
    Pred = LHSUnsigned ? ICmpInst::ICMP_UGE : ICmpInst::ICMP_SGE;
    break;
  case FCmpInst::FCMP_ULT:
  case FCmpInst::FCMP_OLT:
    Pred = LHSUnsigned ? ICmpInst::ICMP_ULT : ICmpInst::ICMP_SLT;
    break;
  case FCmpInst::FCMP_ULE:
  case FCmpInst::FCMP_OLE:
    Pred = LHSUnsigned ? ICmpInst::ICMP_ULE : ICmpInst::ICMP_SLE;
    break;
  case FCmpInst::FCMP_UNE:
  case FCmpInst::FCMP_ONE:
    Pred = ICmpInst::ICMP_NE;
    break;
  case FCmpInst::FCMP_ORD:
    return ReplaceInstUsesWith(I, Builder->getTrue());
  case FCmpInst::FCMP_UNO:
    return ReplaceInstUsesWith(I, Builder->getFalse());
  }

  IntegerType *IntTy = cast<IntegerType>(LHSI->getOperand(0)->getType());

  // Now we know that the APFloat is a normal number, zero or inf.

  // See if the FP constant is too large for the integer.  For example,
  // comparing an i8 to 300.0.
  unsigned IntWidth = IntTy->getScalarSizeInBits();

  if (!LHSUnsigned) {
    // If the RHS value is > SignedMax, fold the comparison.  This handles +INF
    // and large values.
    APFloat SMax(RHS.getSemantics());
    SMax.convertFromAPInt(APInt::getSignedMaxValue(IntWidth), true,
                          APFloat::rmNearestTiesToEven);
    if (SMax.compare(RHS) == APFloat::cmpLessThan) {  // smax < 13123.0
      if (Pred == ICmpInst::ICMP_NE  || Pred == ICmpInst::ICMP_SLT ||
          Pred == ICmpInst::ICMP_SLE)
        return ReplaceInstUsesWith(I, Builder->getTrue());
      return ReplaceInstUsesWith(I, Builder->getFalse());
    }
  } else {
    // If the RHS value is > UnsignedMax, fold the comparison. This handles
    // +INF and large values.
    APFloat UMax(RHS.getSemantics());
    UMax.convertFromAPInt(APInt::getMaxValue(IntWidth), false,
                          APFloat::rmNearestTiesToEven);
    if (UMax.compare(RHS) == APFloat::cmpLessThan) {  // umax < 13123.0
      if (Pred == ICmpInst::ICMP_NE  || Pred == ICmpInst::ICMP_ULT ||
          Pred == ICmpInst::ICMP_ULE)
        return ReplaceInstUsesWith(I, Builder->getTrue());
      return ReplaceInstUsesWith(I, Builder->getFalse());
    }
  }

  if (!LHSUnsigned) {
    // See if the RHS value is < SignedMin.
    APFloat SMin(RHS.getSemantics());
    SMin.convertFromAPInt(APInt::getSignedMinValue(IntWidth), true,
                          APFloat::rmNearestTiesToEven);
    if (SMin.compare(RHS) == APFloat::cmpGreaterThan) { // smin > 12312.0
      if (Pred == ICmpInst::ICMP_NE || Pred == ICmpInst::ICMP_SGT ||
          Pred == ICmpInst::ICMP_SGE)
        return ReplaceInstUsesWith(I, Builder->getTrue());
      return ReplaceInstUsesWith(I, Builder->getFalse());
    }
  } else {
    // See if the RHS value is < UnsignedMin.
    APFloat SMin(RHS.getSemantics());
    SMin.convertFromAPInt(APInt::getMinValue(IntWidth), true,
                          APFloat::rmNearestTiesToEven);
    if (SMin.compare(RHS) == APFloat::cmpGreaterThan) { // umin > 12312.0
      if (Pred == ICmpInst::ICMP_NE || Pred == ICmpInst::ICMP_UGT ||
          Pred == ICmpInst::ICMP_UGE)
        return ReplaceInstUsesWith(I, Builder->getTrue());
      return ReplaceInstUsesWith(I, Builder->getFalse());
    }
  }

  // Okay, now we know that the FP constant fits in the range [SMIN, SMAX] or
  // [0, UMAX], but it may still be fractional.  See if it is fractional by
  // casting the FP value to the integer value and back, checking for equality.
  // Don't do this for zero, because -0.0 is not fractional.
  Constant *RHSInt = LHSUnsigned
    ? ConstantExpr::getFPToUI(RHSC, IntTy)
    : ConstantExpr::getFPToSI(RHSC, IntTy);
  if (!RHS.isZero()) {
    bool Equal = LHSUnsigned
      ? ConstantExpr::getUIToFP(RHSInt, RHSC->getType()) == RHSC
      : ConstantExpr::getSIToFP(RHSInt, RHSC->getType()) == RHSC;
    if (!Equal) {
      // If we had a comparison against a fractional value, we have to adjust
      // the compare predicate and sometimes the value.  RHSC is rounded towards
      // zero at this point.
      switch (Pred) {
      default: llvm_unreachable("Unexpected integer comparison!");
      case ICmpInst::ICMP_NE:  // (float)int != 4.4   --> true
        return ReplaceInstUsesWith(I, Builder->getTrue());
      case ICmpInst::ICMP_EQ:  // (float)int == 4.4   --> false
        return ReplaceInstUsesWith(I, Builder->getFalse());
      case ICmpInst::ICMP_ULE:
        // (float)int <= 4.4   --> int <= 4
        // (float)int <= -4.4  --> false
        if (RHS.isNegative())
          return ReplaceInstUsesWith(I, Builder->getFalse());
        break;
      case ICmpInst::ICMP_SLE:
        // (float)int <= 4.4   --> int <= 4
        // (float)int <= -4.4  --> int < -4
        if (RHS.isNegative())
          Pred = ICmpInst::ICMP_SLT;
        break;
      case ICmpInst::ICMP_ULT:
        // (float)int < -4.4   --> false
        // (float)int < 4.4    --> int <= 4
        if (RHS.isNegative())
          return ReplaceInstUsesWith(I, Builder->getFalse());
        Pred = ICmpInst::ICMP_ULE;
        break;
      case ICmpInst::ICMP_SLT:
        // (float)int < -4.4   --> int < -4
        // (float)int < 4.4    --> int <= 4
        if (!RHS.isNegative())
          Pred = ICmpInst::ICMP_SLE;
        break;
      case ICmpInst::ICMP_UGT:
        // (float)int > 4.4    --> int > 4
        // (float)int > -4.4   --> true
        if (RHS.isNegative())
          return ReplaceInstUsesWith(I, Builder->getTrue());
        break;
      case ICmpInst::ICMP_SGT:
        // (float)int > 4.4    --> int > 4
        // (float)int > -4.4   --> int >= -4
        if (RHS.isNegative())
          Pred = ICmpInst::ICMP_SGE;
        break;
      case ICmpInst::ICMP_UGE:
        // (float)int >= -4.4   --> true
        // (float)int >= 4.4    --> int > 4
        if (RHS.isNegative())
          return ReplaceInstUsesWith(I, Builder->getTrue());
        Pred = ICmpInst::ICMP_UGT;
        break;
      case ICmpInst::ICMP_SGE:
        // (float)int >= -4.4   --> int >= -4
        // (float)int >= 4.4    --> int > 4
        if (!RHS.isNegative())
          Pred = ICmpInst::ICMP_SGT;
        break;
      }
    }
  }

  // Lower this FP comparison into an appropriate integer version of the
  // comparison.
  return new ICmpInst(Pred, LHSI->getOperand(0), RHSInt);
}

Value * InstCombiner::foldFMulConst	(	Instruction *	FMulOrDiv,
		Constant *	C,
		Instruction *	InsertBefore
	)

foldFMulConst() is a helper routine of InstCombiner::visitFMul(). The input FMulOrDiv is a FMul/FDiv with one and only one operand being a constant (i.e. isFMulOrFDivWithConstant(FMulOrDiv) == true). This function is to simplify "FMulOrDiv * C" and returns the resulting expression. Note that this function could return NULL in case the constants cannot be folded into a normal floating-point.

Definition at line 385 of file InstCombineMulDivRem.cpp.

                                                              {
  assert(isFMulOrFDivWithConstant(FMulOrDiv) && "V is invalid");

  Value *Opnd0 = FMulOrDiv->getOperand(0);
  Value *Opnd1 = FMulOrDiv->getOperand(1);

  Constant *C0 = dyn_cast<Constant>(Opnd0);
  Constant *C1 = dyn_cast<Constant>(Opnd1);

  BinaryOperator *R = nullptr;

  // (X * C0) * C => X * (C0*C)
  if (FMulOrDiv->getOpcode() == Instruction::FMul) {
    Constant *F = ConstantExpr::getFMul(C1 ? C1 : C0, C);
    if (isNormalFp(F))
      R = BinaryOperator::CreateFMul(C1 ? Opnd0 : Opnd1, F);
  } else {
    if (C0) {
      // (C0 / X) * C => (C0 * C) / X
      if (FMulOrDiv->hasOneUse()) {
        // It would otherwise introduce another div.
        Constant *F = ConstantExpr::getFMul(C0, C);
        if (isNormalFp(F))
          R = BinaryOperator::CreateFDiv(F, Opnd1);
      }
    } else {
      // (X / C1) * C => X * (C/C1) if C/C1 is not a denormal
      Constant *F = ConstantExpr::getFDiv(C, C1);
      if (isNormalFp(F)) {
        R = BinaryOperator::CreateFMul(Opnd0, F);
      } else {
        // (X / C1) * C => X / (C1/C)
        Constant *F = ConstantExpr::getFDiv(C1, C);
        if (isNormalFp(F))
          R = BinaryOperator::CreateFDiv(Opnd0, F);
      }
    }
  }

  if (R) {
    R->setHasUnsafeAlgebra(true);
    InsertNewInstWith(R, *InsertBefore);
  }

  return R;
}

Instruction * InstCombiner::FoldGEPICmp	(	GEPOperator *	GEPLHS,
		Value *	RHS,
		ICmpInst::Predicate	Cond,
		Instruction &	I
	)

FoldGEPICmp - Fold comparisons between a GEP instruction and something else. At this point we know that the GEP is on the LHS of the comparison.

Definition at line 604 of file InstCombineCompares.cpp.

                                                       {
  // Don't transform signed compares of GEPs into index compares. Even if the
  // GEP is inbounds, the final add of the base pointer can have signed overflow
  // and would change the result of the icmp.
  // e.g. "&foo[0] <s &foo[1]" can't be folded to "true" because "foo" could be
  // the maximum signed value for the pointer type.
  if (ICmpInst::isSigned(Cond))
    return nullptr;

  // Look through bitcasts and addrspacecasts. We do not however want to remove
  // 0 GEPs.
  if (!isa<GetElementPtrInst>(RHS))
    RHS = RHS->stripPointerCasts();

  Value *PtrBase = GEPLHS->getOperand(0);
  if (DL && PtrBase == RHS && GEPLHS->isInBounds()) {
    // ((gep Ptr, OFFSET) cmp Ptr)   ---> (OFFSET cmp 0).
    // This transformation (ignoring the base and scales) is valid because we
    // know pointers can't overflow since the gep is inbounds.  See if we can
    // output an optimized form.
    Value *Offset = EvaluateGEPOffsetExpression(GEPLHS, *this);

    // If not, synthesize the offset the hard way.
    if (!Offset)
      Offset = EmitGEPOffset(GEPLHS);
    return new ICmpInst(ICmpInst::getSignedPredicate(Cond), Offset,
                        Constant::getNullValue(Offset->getType()));
  } else if (GEPOperator *GEPRHS = dyn_cast<GEPOperator>(RHS)) {
    // If the base pointers are different, but the indices are the same, just
    // compare the base pointer.
    if (PtrBase != GEPRHS->getOperand(0)) {
      bool IndicesTheSame = GEPLHS->getNumOperands()==GEPRHS->getNumOperands();
      IndicesTheSame &= GEPLHS->getOperand(0)->getType() ==
                        GEPRHS->getOperand(0)->getType();
      if (IndicesTheSame)
        for (unsigned i = 1, e = GEPLHS->getNumOperands(); i != e; ++i)
          if (GEPLHS->getOperand(i) != GEPRHS->getOperand(i)) {
            IndicesTheSame = false;
            break;
          }

      // If all indices are the same, just compare the base pointers.
      if (IndicesTheSame)
        return new ICmpInst(Cond, GEPLHS->getOperand(0), GEPRHS->getOperand(0));

      // If we're comparing GEPs with two base pointers that only differ in type
      // and both GEPs have only constant indices or just one use, then fold
      // the compare with the adjusted indices.
      if (DL && GEPLHS->isInBounds() && GEPRHS->isInBounds() &&
          (GEPLHS->hasAllConstantIndices() || GEPLHS->hasOneUse()) &&
          (GEPRHS->hasAllConstantIndices() || GEPRHS->hasOneUse()) &&
          PtrBase->stripPointerCasts() ==
            GEPRHS->getOperand(0)->stripPointerCasts()) {
        Value *LOffset = EmitGEPOffset(GEPLHS);
        Value *ROffset = EmitGEPOffset(GEPRHS);

        // If we looked through an addrspacecast between different sized address
        // spaces, the LHS and RHS pointers are different sized
        // integers. Truncate to the smaller one.
        Type *LHSIndexTy = LOffset->getType();
        Type *RHSIndexTy = ROffset->getType();
        if (LHSIndexTy != RHSIndexTy) {
          if (LHSIndexTy->getPrimitiveSizeInBits() <
              RHSIndexTy->getPrimitiveSizeInBits()) {
            ROffset = Builder->CreateTrunc(ROffset, LHSIndexTy);
          } else
            LOffset = Builder->CreateTrunc(LOffset, RHSIndexTy);
        }

        Value *Cmp = Builder->CreateICmp(ICmpInst::getSignedPredicate(Cond),
                                         LOffset, ROffset);
        return ReplaceInstUsesWith(I, Cmp);
      }

      // Otherwise, the base pointers are different and the indices are
      // different, bail out.
      return nullptr;
    }

    // If one of the GEPs has all zero indices, recurse.
    if (GEPLHS->hasAllZeroIndices())
      return FoldGEPICmp(GEPRHS, GEPLHS->getOperand(0),
                         ICmpInst::getSwappedPredicate(Cond), I);

    // If the other GEP has all zero indices, recurse.
    if (GEPRHS->hasAllZeroIndices())
      return FoldGEPICmp(GEPLHS, GEPRHS->getOperand(0), Cond, I);

    bool GEPsInBounds = GEPLHS->isInBounds() && GEPRHS->isInBounds();
    if (GEPLHS->getNumOperands() == GEPRHS->getNumOperands()) {
      // If the GEPs only differ by one index, compare it.
      unsigned NumDifferences = 0;  // Keep track of # differences.
      unsigned DiffOperand = 0;     // The operand that differs.
      for (unsigned i = 1, e = GEPRHS->getNumOperands(); i != e; ++i)
        if (GEPLHS->getOperand(i) != GEPRHS->getOperand(i)) {
          if (GEPLHS->getOperand(i)->getType()->getPrimitiveSizeInBits() !=
                   GEPRHS->getOperand(i)->getType()->getPrimitiveSizeInBits()) {
            // Irreconcilable differences.
            NumDifferences = 2;
            break;
          } else {
            if (NumDifferences++) break;
            DiffOperand = i;
          }
        }

      if (NumDifferences == 0)   // SAME GEP?
        return ReplaceInstUsesWith(I, // No comparison is needed here.
                             Builder->getInt1(ICmpInst::isTrueWhenEqual(Cond)));

      else if (NumDifferences == 1 && GEPsInBounds) {
        Value *LHSV = GEPLHS->getOperand(DiffOperand);
        Value *RHSV = GEPRHS->getOperand(DiffOperand);
        // Make sure we do a signed comparison here.
        return new ICmpInst(ICmpInst::getSignedPredicate(Cond), LHSV, RHSV);
      }
    }

    // Only lower this if the icmp is the only user of the GEP or if we expect
    // the result to fold to a constant!
    if (DL &&
        GEPsInBounds &&
        (isa<ConstantExpr>(GEPLHS) || GEPLHS->hasOneUse()) &&
        (isa<ConstantExpr>(GEPRHS) || GEPRHS->hasOneUse())) {
      // ((gep Ptr, OFFSET1) cmp (gep Ptr, OFFSET2)  --->  (OFFSET1 cmp OFFSET2)
      Value *L = EmitGEPOffset(GEPLHS);
      Value *R = EmitGEPOffset(GEPRHS);
      return new ICmpInst(ICmpInst::getSignedPredicate(Cond), L, R);
    }
  }
  return nullptr;
}

Instruction * InstCombiner::FoldICmpAddOpCst	(	Instruction &	ICI,
		Value *	X,
		ConstantInt *	CI,
		ICmpInst::Predicate	Pred
	)

FoldICmpAddOpCst - Fold "icmp pred (X+CI), X".

Definition at line 740 of file InstCombineCompares.cpp.

                                                                    {
  // If we have X+0, exit early (simplifying logic below) and let it get folded
  // elsewhere.   icmp X+0, X  -> icmp X, X
  if (CI->isZero()) {
    bool isTrue = ICmpInst::isTrueWhenEqual(Pred);
    return ReplaceInstUsesWith(ICI, ConstantInt::get(ICI.getType(), isTrue));
  }

  // (X+4) == X -> false.
  if (Pred == ICmpInst::ICMP_EQ)
    return ReplaceInstUsesWith(ICI, Builder->getFalse());

  // (X+4) != X -> true.
  if (Pred == ICmpInst::ICMP_NE)
    return ReplaceInstUsesWith(ICI, Builder->getTrue());

  // From this point on, we know that (X+C <= X) --> (X+C < X) because C != 0,
  // so the values can never be equal.  Similarly for all other "or equals"
  // operators.

  // (X+1) <u X        --> X >u (MAXUINT-1)        --> X == 255
  // (X+2) <u X        --> X >u (MAXUINT-2)        --> X > 253
  // (X+MAXUINT) <u X  --> X >u (MAXUINT-MAXUINT)  --> X != 0
  if (Pred == ICmpInst::ICMP_ULT || Pred == ICmpInst::ICMP_ULE) {
    Value *R =
      ConstantExpr::getSub(ConstantInt::getAllOnesValue(CI->getType()), CI);
    return new ICmpInst(ICmpInst::ICMP_UGT, X, R);
  }

  // (X+1) >u X        --> X <u (0-1)        --> X != 255
  // (X+2) >u X        --> X <u (0-2)        --> X <u 254
  // (X+MAXUINT) >u X  --> X <u (0-MAXUINT)  --> X <u 1  --> X == 0
  if (Pred == ICmpInst::ICMP_UGT || Pred == ICmpInst::ICMP_UGE)
    return new ICmpInst(ICmpInst::ICMP_ULT, X, ConstantExpr::getNeg(CI));

  unsigned BitWidth = CI->getType()->getPrimitiveSizeInBits();
  ConstantInt *SMax = ConstantInt::get(X->getContext(),
                                       APInt::getSignedMaxValue(BitWidth));

  // (X+ 1) <s X       --> X >s (MAXSINT-1)          --> X == 127
  // (X+ 2) <s X       --> X >s (MAXSINT-2)          --> X >s 125
  // (X+MAXSINT) <s X  --> X >s (MAXSINT-MAXSINT)    --> X >s 0
  // (X+MINSINT) <s X  --> X >s (MAXSINT-MINSINT)    --> X >s -1
  // (X+ -2) <s X      --> X >s (MAXSINT- -2)        --> X >s 126
  // (X+ -1) <s X      --> X >s (MAXSINT- -1)        --> X != 127
  if (Pred == ICmpInst::ICMP_SLT || Pred == ICmpInst::ICMP_SLE)
    return new ICmpInst(ICmpInst::ICMP_SGT, X, ConstantExpr::getSub(SMax, CI));

  // (X+ 1) >s X       --> X <s (MAXSINT-(1-1))       --> X != 127
  // (X+ 2) >s X       --> X <s (MAXSINT-(2-1))       --> X <s 126
  // (X+MAXSINT) >s X  --> X <s (MAXSINT-(MAXSINT-1)) --> X <s 1
  // (X+MINSINT) >s X  --> X <s (MAXSINT-(MINSINT-1)) --> X <s -2
  // (X+ -2) >s X      --> X <s (MAXSINT-(-2-1))      --> X <s -126
  // (X+ -1) >s X      --> X <s (MAXSINT-(-1-1))      --> X == -128

  assert(Pred == ICmpInst::ICMP_SGT || Pred == ICmpInst::ICMP_SGE);
  Constant *C = Builder->getInt(CI->getValue()-1);
  return new ICmpInst(ICmpInst::ICMP_SLT, X, ConstantExpr::getSub(SMax, C));
}

Instruction * InstCombiner::FoldICmpDivCst	(	ICmpInst &	ICI,
		BinaryOperator *	DivI,
		ConstantInt *	DivRHS
	)

FoldICmpDivCst - Fold "icmp pred, ([su]div X, DivRHS), CmpRHS" where DivRHS and CmpRHS are both known to be integer constants.

If the division is known to be exact, then there is no remainder from the divide, so the covered range size is unit, otherwise it is the divisor.

Definition at line 804 of file InstCombineCompares.cpp.

                                                               {
  ConstantInt *CmpRHS = cast<ConstantInt>(ICI.getOperand(1));
  const APInt &CmpRHSV = CmpRHS->getValue();

  // FIXME: If the operand types don't match the type of the divide
  // then don't attempt this transform. The code below doesn't have the
  // logic to deal with a signed divide and an unsigned compare (and
  // vice versa). This is because (x /s C1) <s C2  produces different
  // results than (x /s C1) <u C2 or (x /u C1) <s C2 or even
  // (x /u C1) <u C2.  Simply casting the operands and result won't
  // work. :(  The if statement below tests that condition and bails
  // if it finds it.
  bool DivIsSigned = DivI->getOpcode() == Instruction::SDiv;
  if (!ICI.isEquality() && DivIsSigned != ICI.isSigned())
    return nullptr;
  if (DivRHS->isZero())
    return nullptr; // The ProdOV computation fails on divide by zero.
  if (DivIsSigned && DivRHS->isAllOnesValue())
    return nullptr; // The overflow computation also screws up here
  if (DivRHS->isOne()) {
    // This eliminates some funny cases with INT_MIN.
    ICI.setOperand(0, DivI->getOperand(0));   // X/1 == X.
    return &ICI;
  }

  // Compute Prod = CI * DivRHS. We are essentially solving an equation
  // of form X/C1=C2. We solve for X by multiplying C1 (DivRHS) and
  // C2 (CI). By solving for X we can turn this into a range check
  // instead of computing a divide.
  Constant *Prod = ConstantExpr::getMul(CmpRHS, DivRHS);

  // Determine if the product overflows by seeing if the product is
  // not equal to the divide. Make sure we do the same kind of divide
  // as in the LHS instruction that we're folding.
  bool ProdOV = (DivIsSigned ? ConstantExpr::getSDiv(Prod, DivRHS) :
                 ConstantExpr::getUDiv(Prod, DivRHS)) != CmpRHS;

  // Get the ICmp opcode
  ICmpInst::Predicate Pred = ICI.getPredicate();

  /// If the division is known to be exact, then there is no remainder from the
  /// divide, so the covered range size is unit, otherwise it is the divisor.
  ConstantInt *RangeSize = DivI->isExact() ? getOne(Prod) : DivRHS;

  // Figure out the interval that is being checked.  For example, a comparison
  // like "X /u 5 == 0" is really checking that X is in the interval [0, 5).
  // Compute this interval based on the constants involved and the signedness of
  // the compare/divide.  This computes a half-open interval, keeping track of
  // whether either value in the interval overflows.  After analysis each
  // overflow variable is set to 0 if it's corresponding bound variable is valid
  // -1 if overflowed off the bottom end, or +1 if overflowed off the top end.
  int LoOverflow = 0, HiOverflow = 0;
  Constant *LoBound = nullptr, *HiBound = nullptr;

  if (!DivIsSigned) {  // udiv
    // e.g. X/5 op 3  --> [15, 20)
    LoBound = Prod;
    HiOverflow = LoOverflow = ProdOV;
    if (!HiOverflow) {
      // If this is not an exact divide, then many values in the range collapse
      // to the same result value.
      HiOverflow = AddWithOverflow(HiBound, LoBound, RangeSize, false);
    }

  } else if (DivRHS->getValue().isStrictlyPositive()) { // Divisor is > 0.
    if (CmpRHSV == 0) {       // (X / pos) op 0
      // Can't overflow.  e.g.  X/2 op 0 --> [-1, 2)
      LoBound = ConstantExpr::getNeg(SubOne(RangeSize));
      HiBound = RangeSize;
    } else if (CmpRHSV.isStrictlyPositive()) {   // (X / pos) op pos
      LoBound = Prod;     // e.g.   X/5 op 3 --> [15, 20)
      HiOverflow = LoOverflow = ProdOV;
      if (!HiOverflow)
        HiOverflow = AddWithOverflow(HiBound, Prod, RangeSize, true);
    } else {                       // (X / pos) op neg
      // e.g. X/5 op -3  --> [-15-4, -15+1) --> [-19, -14)
      HiBound = AddOne(Prod);
      LoOverflow = HiOverflow = ProdOV ? -1 : 0;
      if (!LoOverflow) {
        ConstantInt *DivNeg =cast<ConstantInt>(ConstantExpr::getNeg(RangeSize));
        LoOverflow = AddWithOverflow(LoBound, HiBound, DivNeg, true) ? -1 : 0;
      }
    }
  } else if (DivRHS->isNegative()) { // Divisor is < 0.
    if (DivI->isExact())
      RangeSize = cast<ConstantInt>(ConstantExpr::getNeg(RangeSize));
    if (CmpRHSV == 0) {       // (X / neg) op 0
      // e.g. X/-5 op 0  --> [-4, 5)
      LoBound = AddOne(RangeSize);
      HiBound = cast<ConstantInt>(ConstantExpr::getNeg(RangeSize));
      if (HiBound == DivRHS) {     // -INTMIN = INTMIN
        HiOverflow = 1;            // [INTMIN+1, overflow)
        HiBound = nullptr;         // e.g. X/INTMIN = 0 --> X > INTMIN
      }
    } else if (CmpRHSV.isStrictlyPositive()) {   // (X / neg) op pos
      // e.g. X/-5 op 3  --> [-19, -14)
      HiBound = AddOne(Prod);
      HiOverflow = LoOverflow = ProdOV ? -1 : 0;
      if (!LoOverflow)
        LoOverflow = AddWithOverflow(LoBound, HiBound, RangeSize, true) ? -1:0;
    } else {                       // (X / neg) op neg
      LoBound = Prod;       // e.g. X/-5 op -3  --> [15, 20)
      LoOverflow = HiOverflow = ProdOV;
      if (!HiOverflow)
        HiOverflow = SubWithOverflow(HiBound, Prod, RangeSize, true);
    }

    // Dividing by a negative swaps the condition.  LT <-> GT
    Pred = ICmpInst::getSwappedPredicate(Pred);
  }

  Value *X = DivI->getOperand(0);
  switch (Pred) {
  default: llvm_unreachable("Unhandled icmp opcode!");
  case ICmpInst::ICMP_EQ:
    if (LoOverflow && HiOverflow)
      return ReplaceInstUsesWith(ICI, Builder->getFalse());
    if (HiOverflow)
      return new ICmpInst(DivIsSigned ? ICmpInst::ICMP_SGE :
                          ICmpInst::ICMP_UGE, X, LoBound);
    if (LoOverflow)
      return new ICmpInst(DivIsSigned ? ICmpInst::ICMP_SLT :
                          ICmpInst::ICMP_ULT, X, HiBound);
    return ReplaceInstUsesWith(ICI, InsertRangeTest(X, LoBound, HiBound,
                                                    DivIsSigned, true));
  case ICmpInst::ICMP_NE:
    if (LoOverflow && HiOverflow)
      return ReplaceInstUsesWith(ICI, Builder->getTrue());
    if (HiOverflow)
      return new ICmpInst(DivIsSigned ? ICmpInst::ICMP_SLT :
                          ICmpInst::ICMP_ULT, X, LoBound);
    if (LoOverflow)
      return new ICmpInst(DivIsSigned ? ICmpInst::ICMP_SGE :
                          ICmpInst::ICMP_UGE, X, HiBound);
    return ReplaceInstUsesWith(ICI, InsertRangeTest(X, LoBound, HiBound,
                                                    DivIsSigned, false));
  case ICmpInst::ICMP_ULT:
  case ICmpInst::ICMP_SLT:
    if (LoOverflow == +1)   // Low bound is greater than input range.
      return ReplaceInstUsesWith(ICI, Builder->getTrue());
    if (LoOverflow == -1)   // Low bound is less than input range.
      return ReplaceInstUsesWith(ICI, Builder->getFalse());
    return new ICmpInst(Pred, X, LoBound);
  case ICmpInst::ICMP_UGT:
  case ICmpInst::ICMP_SGT:
    if (HiOverflow == +1)       // High bound greater than input range.
      return ReplaceInstUsesWith(ICI, Builder->getFalse());
    if (HiOverflow == -1)       // High bound less than input range.
      return ReplaceInstUsesWith(ICI, Builder->getTrue());
    if (Pred == ICmpInst::ICMP_UGT)
      return new ICmpInst(ICmpInst::ICMP_UGE, X, HiBound);
    return new ICmpInst(ICmpInst::ICMP_SGE, X, HiBound);
  }
}

Instruction * InstCombiner::FoldICmpShrCst	(	ICmpInst &	ICI,
		BinaryOperator *	DivI,
		ConstantInt *	DivRHS
	)

FoldICmpShrCst - Handle "icmp(([al]shr X, cst1), cst2)".

Definition at line 961 of file InstCombineCompares.cpp.

                                                              {
  const APInt &CmpRHSV = cast<ConstantInt>(ICI.getOperand(1))->getValue();

  // Check that the shift amount is in range.  If not, don't perform
  // undefined shifts.  When the shift is visited it will be
  // simplified.
  uint32_t TypeBits = CmpRHSV.getBitWidth();
  uint32_t ShAmtVal = (uint32_t)ShAmt->getLimitedValue(TypeBits);
  if (ShAmtVal >= TypeBits || ShAmtVal == 0)
    return nullptr;

  if (!ICI.isEquality()) {
    // If we have an unsigned comparison and an ashr, we can't simplify this.
    // Similarly for signed comparisons with lshr.
    if (ICI.isSigned() != (Shr->getOpcode() == Instruction::AShr))
      return nullptr;

    // Otherwise, all lshr and most exact ashr's are equivalent to a udiv/sdiv
    // by a power of 2.  Since we already have logic to simplify these,
    // transform to div and then simplify the resultant comparison.
    if (Shr->getOpcode() == Instruction::AShr &&
        (!Shr->isExact() || ShAmtVal == TypeBits - 1))
      return nullptr;

    // Revisit the shift (to delete it).
    Worklist.Add(Shr);

    Constant *DivCst =
      ConstantInt::get(Shr->getType(), APInt::getOneBitSet(TypeBits, ShAmtVal));

    Value *Tmp =
      Shr->getOpcode() == Instruction::AShr ?
      Builder->CreateSDiv(Shr->getOperand(0), DivCst, "", Shr->isExact()) :
      Builder->CreateUDiv(Shr->getOperand(0), DivCst, "", Shr->isExact());

    ICI.setOperand(0, Tmp);

    // If the builder folded the binop, just return it.
    BinaryOperator *TheDiv = dyn_cast<BinaryOperator>(Tmp);
    if (!TheDiv)
      return &ICI;

    // Otherwise, fold this div/compare.
    assert(TheDiv->getOpcode() == Instruction::SDiv ||
           TheDiv->getOpcode() == Instruction::UDiv);

    Instruction *Res = FoldICmpDivCst(ICI, TheDiv, cast<ConstantInt>(DivCst));
    assert(Res && "This div/cst should have folded!");
    return Res;
  }


  // If we are comparing against bits always shifted out, the
  // comparison cannot succeed.
  APInt Comp = CmpRHSV << ShAmtVal;
  ConstantInt *ShiftedCmpRHS = Builder->getInt(Comp);
  if (Shr->getOpcode() == Instruction::LShr)
    Comp = Comp.lshr(ShAmtVal);
  else
    Comp = Comp.ashr(ShAmtVal);

  if (Comp != CmpRHSV) { // Comparing against a bit that we know is zero.
    bool IsICMP_NE = ICI.getPredicate() == ICmpInst::ICMP_NE;
    Constant *Cst = Builder->getInt1(IsICMP_NE);
    return ReplaceInstUsesWith(ICI, Cst);
  }

  // Otherwise, check to see if the bits shifted out are known to be zero.
  // If so, we can compare against the unshifted value:
  //  (X & 4) >> 1 == 2  --> (X & 4) == 4.
  if (Shr->hasOneUse() && Shr->isExact())
    return new ICmpInst(ICI.getPredicate(), Shr->getOperand(0), ShiftedCmpRHS);

  if (Shr->hasOneUse()) {
    // Otherwise strength reduce the shift into an and.
    APInt Val(APInt::getHighBitsSet(TypeBits, TypeBits - ShAmtVal));
    Constant *Mask = Builder->getInt(Val);

    Value *And = Builder->CreateAnd(Shr->getOperand(0),
                                    Mask, Shr->getName()+".mask");
    return new ICmpInst(ICI.getPredicate(), And, ShiftedCmpRHS);
  }
  return nullptr;
}

Value * InstCombiner::FoldLogicalPlusAnd ( Value *  LHS, Value *  RHS, ConstantInt *  Mask, bool  isSub, Instruction &  I  )

private

FoldLogicalPlusAnd - This is part of an expression (LHS +/- RHS) & Mask, where isSub determines whether the operator is a sub. If we can fold one of the following xforms:

((A & N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == Mask ((A | N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == 0 ((A ^ N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == 0

return (A +/- B).

Definition at line 332 of file InstCombineAndOrXor.cpp.

                                                        {
  Instruction *LHSI = dyn_cast<Instruction>(LHS);
  if (!LHSI || LHSI->getNumOperands() != 2 ||
      !isa<ConstantInt>(LHSI->getOperand(1))) return nullptr;

  ConstantInt *N = cast<ConstantInt>(LHSI->getOperand(1));

  switch (LHSI->getOpcode()) {
  default: return nullptr;
  case Instruction::And:
    if (ConstantExpr::getAnd(N, Mask) == Mask) {
      // If the AndRHS is a power of two minus one (0+1+), this is simple.
      if ((Mask->getValue().countLeadingZeros() +
           Mask->getValue().countPopulation()) ==
          Mask->getValue().getBitWidth())
        break;

      // Otherwise, if Mask is 0+1+0+, and if B is known to have the low 0+
      // part, we don't need any explicit masks to take them out of A.  If that
      // is all N is, ignore it.
      uint32_t MB = 0, ME = 0;
      if (isRunOfOnes(Mask, MB, ME)) {  // begin/end bit of run, inclusive
        uint32_t BitWidth = cast<IntegerType>(RHS->getType())->getBitWidth();
        APInt Mask(APInt::getLowBitsSet(BitWidth, MB-1));
        if (MaskedValueIsZero(RHS, Mask))
          break;
      }
    }
    return nullptr;
  case Instruction::Or:
  case Instruction::Xor:
    // If the AndRHS is a power of two minus one (0+1+), and N&Mask == 0
    if ((Mask->getValue().countLeadingZeros() +
         Mask->getValue().countPopulation()) == Mask->getValue().getBitWidth()
        && ConstantExpr::getAnd(N, Mask)->isNullValue())
      break;
    return nullptr;
  }

  if (isSub)
    return Builder->CreateSub(LHSI->getOperand(0), RHS, "fold");
  return Builder->CreateAdd(LHSI->getOperand(0), RHS, "fold");
}

Instruction * InstCombiner::FoldOpIntoPhi ( Instruction &  I )

private

FoldOpIntoPhi - Given a binary operator, cast instruction, or select which has a PHI node as operand #0, see if we can fold the instruction into the PHI (which is only possible if all operands to the PHI are constants).

Definition at line 724 of file InstructionCombining.cpp.

                                                       {
  PHINode *PN = cast<PHINode>(I.getOperand(0));
  unsigned NumPHIValues = PN->getNumIncomingValues();
  if (NumPHIValues == 0)
    return nullptr;

  // We normally only transform phis with a single use.  However, if a PHI has
  // multiple uses and they are all the same operation, we can fold *all* of the
  // uses into the PHI.
  if (!PN->hasOneUse()) {
    // Walk the use list for the instruction, comparing them to I.
    for (User *U : PN->users()) {
      Instruction *UI = cast<Instruction>(U);
      if (UI != &I && !I.isIdenticalTo(UI))
        return nullptr;
    }
    // Otherwise, we can replace *all* users with the new PHI we form.
  }

  // Check to see if all of the operands of the PHI are simple constants
  // (constantint/constantfp/undef).  If there is one non-constant value,
  // remember the BB it is in.  If there is more than one or if *it* is a PHI,
  // bail out.  We don't do arbitrary constant expressions here because moving
  // their computation can be expensive without a cost model.
  BasicBlock *NonConstBB = nullptr;
  for (unsigned i = 0; i != NumPHIValues; ++i) {
    Value *InVal = PN->getIncomingValue(i);
    if (isa<Constant>(InVal) && !isa<ConstantExpr>(InVal))
      continue;

    if (isa<PHINode>(InVal)) return nullptr;  // Itself a phi.
    if (NonConstBB) return nullptr;  // More than one non-const value.

    NonConstBB = PN->getIncomingBlock(i);

    // If the InVal is an invoke at the end of the pred block, then we can't
    // insert a computation after it without breaking the edge.
    if (InvokeInst *II = dyn_cast<InvokeInst>(InVal))
      if (II->getParent() == NonConstBB)
        return nullptr;

    // If the incoming non-constant value is in I's block, we will remove one
    // instruction, but insert another equivalent one, leading to infinite
    // instcombine.
    if (NonConstBB == I.getParent())
      return nullptr;
  }

  // If there is exactly one non-constant value, we can insert a copy of the
  // operation in that block.  However, if this is a critical edge, we would be
  // inserting the computation one some other paths (e.g. inside a loop).  Only
  // do this if the pred block is unconditionally branching into the phi block.
  if (NonConstBB != nullptr) {
    BranchInst *BI = dyn_cast<BranchInst>(NonConstBB->getTerminator());
    if (!BI || !BI->isUnconditional()) return nullptr;
  }

  // Okay, we can do the transformation: create the new PHI node.
  PHINode *NewPN = PHINode::Create(I.getType(), PN->getNumIncomingValues());
  InsertNewInstBefore(NewPN, *PN);
  NewPN->takeName(PN);

  // If we are going to have to insert a new computation, do so right before the
  // predecessors terminator.
  if (NonConstBB)
    Builder->SetInsertPoint(NonConstBB->getTerminator());

  // Next, add all of the operands to the PHI.
  if (SelectInst *SI = dyn_cast<SelectInst>(&I)) {
    // We only currently try to fold the condition of a select when it is a phi,
    // not the true/false values.
    Value *TrueV = SI->getTrueValue();
    Value *FalseV = SI->getFalseValue();
    BasicBlock *PhiTransBB = PN->getParent();
    for (unsigned i = 0; i != NumPHIValues; ++i) {
      BasicBlock *ThisBB = PN->getIncomingBlock(i);
      Value *TrueVInPred = TrueV->DoPHITranslation(PhiTransBB, ThisBB);
      Value *FalseVInPred = FalseV->DoPHITranslation(PhiTransBB, ThisBB);
      Value *InV = nullptr;
      // Beware of ConstantExpr:  it may eventually evaluate to getNullValue,
      // even if currently isNullValue gives false.
      Constant *InC = dyn_cast<Constant>(PN->getIncomingValue(i));
      if (InC && !isa<ConstantExpr>(InC))
        InV = InC->isNullValue() ? FalseVInPred : TrueVInPred;
      else
        InV = Builder->CreateSelect(PN->getIncomingValue(i),
                                    TrueVInPred, FalseVInPred, "phitmp");
      NewPN->addIncoming(InV, ThisBB);
    }
  } else if (CmpInst *CI = dyn_cast<CmpInst>(&I)) {
    Constant *C = cast<Constant>(I.getOperand(1));
    for (unsigned i = 0; i != NumPHIValues; ++i) {
      Value *InV = nullptr;
      if (Constant *InC = dyn_cast<Constant>(PN->getIncomingValue(i)))
        InV = ConstantExpr::getCompare(CI->getPredicate(), InC, C);
      else if (isa<ICmpInst>(CI))
        InV = Builder->CreateICmp(CI->getPredicate(), PN->getIncomingValue(i),
                                  C, "phitmp");
      else
        InV = Builder->CreateFCmp(CI->getPredicate(), PN->getIncomingValue(i),
                                  C, "phitmp");
      NewPN->addIncoming(InV, PN->getIncomingBlock(i));
    }
  } else if (I.getNumOperands() == 2) {
    Constant *C = cast<Constant>(I.getOperand(1));
    for (unsigned i = 0; i != NumPHIValues; ++i) {
      Value *InV = nullptr;
      if (Constant *InC = dyn_cast<Constant>(PN->getIncomingValue(i)))
        InV = ConstantExpr::get(I.getOpcode(), InC, C);
      else
        InV = Builder->CreateBinOp(cast<BinaryOperator>(I).getOpcode(),
                                   PN->getIncomingValue(i), C, "phitmp");
      NewPN->addIncoming(InV, PN->getIncomingBlock(i));
    }
  } else {
    CastInst *CI = cast<CastInst>(&I);
    Type *RetTy = CI->getType();
    for (unsigned i = 0; i != NumPHIValues; ++i) {
      Value *InV;
      if (Constant *InC = dyn_cast<Constant>(PN->getIncomingValue(i)))
        InV = ConstantExpr::getCast(CI->getOpcode(), InC, RetTy);
      else
        InV = Builder->CreateCast(CI->getOpcode(),
                                PN->getIncomingValue(i), I.getType(), "phitmp");
      NewPN->addIncoming(InV, PN->getIncomingBlock(i));
    }
  }

  for (auto UI = PN->user_begin(), E = PN->user_end(); UI != E;) {
    Instruction *User = cast<Instruction>(*UI++);
    if (User == &I) continue;
    ReplaceInstUsesWith(*User, NewPN);
    EraseInstFromFunction(*User);
  }
  return ReplaceInstUsesWith(I, NewPN);
}

Instruction * InstCombiner::FoldOpIntoSelect ( Instruction &  Op, SelectInst *  SI  )

private

Definition at line 687 of file InstructionCombining.cpp.

                                                                           {
  // Don't modify shared select instructions
  if (!SI->hasOneUse()) return nullptr;
  Value *TV = SI->getOperand(1);
  Value *FV = SI->getOperand(2);

  if (isa<Constant>(TV) || isa<Constant>(FV)) {
    // Bool selects with constant operands can be folded to logical ops.
    if (SI->getType()->isIntegerTy(1)) return nullptr;

    // If it's a bitcast involving vectors, make sure it has the same number of
    // elements on both sides.
    if (BitCastInst *BC = dyn_cast<BitCastInst>(&Op)) {
      VectorType *DestTy = dyn_cast<VectorType>(BC->getDestTy());
      VectorType *SrcTy = dyn_cast<VectorType>(BC->getSrcTy());

      // Verify that either both or neither are vectors.
      if ((SrcTy == nullptr) != (DestTy == nullptr)) return nullptr;
      // If vectors, verify that they have the same number of elements.
      if (SrcTy && SrcTy->getNumElements() != DestTy->getNumElements())
        return nullptr;
    }

    Value *SelectTrueVal = FoldOperationIntoSelectOperand(Op, TV, this);
    Value *SelectFalseVal = FoldOperationIntoSelectOperand(Op, FV, this);

    return SelectInst::Create(SI->getCondition(),
                              SelectTrueVal, SelectFalseVal);
  }
  return nullptr;
}

Value * InstCombiner::FoldOrOfFCmps	(	FCmpInst *	LHS,
		FCmpInst *	RHS
	)

FoldOrOfFCmps - Optimize (fcmp)|(fcmp). NOTE: Unlike the rest of instcombine, this returns a Value which should already be inserted into the function.

Definition at line 1823 of file InstCombineAndOrXor.cpp.

                                                               {
  if (LHS->getPredicate() == FCmpInst::FCMP_UNO &&
      RHS->getPredicate() == FCmpInst::FCMP_UNO &&
      LHS->getOperand(0)->getType() == RHS->getOperand(0)->getType()) {
    if (ConstantFP *LHSC = dyn_cast<ConstantFP>(LHS->getOperand(1)))
      if (ConstantFP *RHSC = dyn_cast<ConstantFP>(RHS->getOperand(1))) {
        // If either of the constants are nans, then the whole thing returns
        // true.
        if (LHSC->getValueAPF().isNaN() || RHSC->getValueAPF().isNaN())
          return Builder->getTrue();

        // Otherwise, no need to compare the two constants, compare the
        // rest.
        return Builder->CreateFCmpUNO(LHS->getOperand(0), RHS->getOperand(0));
      }

    // Handle vector zeros.  This occurs because the canonical form of
    // "fcmp uno x,x" is "fcmp uno x, 0".
    if (isa<ConstantAggregateZero>(LHS->getOperand(1)) &&
        isa<ConstantAggregateZero>(RHS->getOperand(1)))
      return Builder->CreateFCmpUNO(LHS->getOperand(0), RHS->getOperand(0));

    return nullptr;
  }

  Value *Op0LHS = LHS->getOperand(0), *Op0RHS = LHS->getOperand(1);
  Value *Op1LHS = RHS->getOperand(0), *Op1RHS = RHS->getOperand(1);
  FCmpInst::Predicate Op0CC = LHS->getPredicate(), Op1CC = RHS->getPredicate();

  if (Op0LHS == Op1RHS && Op0RHS == Op1LHS) {
    // Swap RHS operands to match LHS.
    Op1CC = FCmpInst::getSwappedPredicate(Op1CC);
    std::swap(Op1LHS, Op1RHS);
  }
  if (Op0LHS == Op1LHS && Op0RHS == Op1RHS) {
    // Simplify (fcmp cc0 x, y) | (fcmp cc1 x, y).
    if (Op0CC == Op1CC)
      return Builder->CreateFCmp((FCmpInst::Predicate)Op0CC, Op0LHS, Op0RHS);
    if (Op0CC == FCmpInst::FCMP_TRUE || Op1CC == FCmpInst::FCMP_TRUE)
      return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 1);
    if (Op0CC == FCmpInst::FCMP_FALSE)
      return RHS;
    if (Op1CC == FCmpInst::FCMP_FALSE)
      return LHS;
    bool Op0Ordered;
    bool Op1Ordered;
    unsigned Op0Pred = getFCmpCode(Op0CC, Op0Ordered);
    unsigned Op1Pred = getFCmpCode(Op1CC, Op1Ordered);
    if (Op0Ordered == Op1Ordered) {
      // If both are ordered or unordered, return a new fcmp with
      // or'ed predicates.
      return getFCmpValue(Op0Ordered, Op0Pred|Op1Pred, Op0LHS, Op0RHS, Builder);
    }
  }
  return nullptr;
}

Value * InstCombiner::FoldOrOfICmps	(	ICmpInst *	LHS,
		ICmpInst *	RHS
	)

FoldOrOfICmps - Fold (icmp)|(icmp) if possible.

Definition at line 1557 of file InstCombineAndOrXor.cpp.

                                                               {
  ICmpInst::Predicate LHSCC = LHS->getPredicate(), RHSCC = RHS->getPredicate();

  // Fold (iszero(A & K1) | iszero(A & K2)) ->  (A & (K1 | K2)) != (K1 | K2)
  // if K1 and K2 are a one-bit mask.
  ConstantInt *LHSCst = dyn_cast<ConstantInt>(LHS->getOperand(1));
  ConstantInt *RHSCst = dyn_cast<ConstantInt>(RHS->getOperand(1));

  if (LHS->getPredicate() == ICmpInst::ICMP_EQ && LHSCst && LHSCst->isZero() &&
      RHS->getPredicate() == ICmpInst::ICMP_EQ && RHSCst && RHSCst->isZero()) {

    BinaryOperator *LAnd = dyn_cast<BinaryOperator>(LHS->getOperand(0));
    BinaryOperator *RAnd = dyn_cast<BinaryOperator>(RHS->getOperand(0));
    if (LAnd && RAnd && LAnd->hasOneUse() && RHS->hasOneUse() &&
        LAnd->getOpcode() == Instruction::And &&
        RAnd->getOpcode() == Instruction::And) {

      Value *Mask = nullptr;
      Value *Masked = nullptr;
      if (LAnd->getOperand(0) == RAnd->getOperand(0) &&
          isKnownToBeAPowerOfTwo(LAnd->getOperand(1)) &&
          isKnownToBeAPowerOfTwo(RAnd->getOperand(1))) {
        Mask = Builder->CreateOr(LAnd->getOperand(1), RAnd->getOperand(1));
        Masked = Builder->CreateAnd(LAnd->getOperand(0), Mask);
      } else if (LAnd->getOperand(1) == RAnd->getOperand(1) &&
                 isKnownToBeAPowerOfTwo(LAnd->getOperand(0)) &&
                 isKnownToBeAPowerOfTwo(RAnd->getOperand(0))) {
        Mask = Builder->CreateOr(LAnd->getOperand(0), RAnd->getOperand(0));
        Masked = Builder->CreateAnd(LAnd->getOperand(1), Mask);
      }

      if (Masked)
        return Builder->CreateICmp(ICmpInst::ICMP_NE, Masked, Mask);
    }
  }

  // (icmp1 A, B) | (icmp2 A, B) --> (icmp3 A, B)
  if (PredicatesFoldable(LHSCC, RHSCC)) {
    if (LHS->getOperand(0) == RHS->getOperand(1) &&
        LHS->getOperand(1) == RHS->getOperand(0))
      LHS->swapOperands();
    if (LHS->getOperand(0) == RHS->getOperand(0) &&
        LHS->getOperand(1) == RHS->getOperand(1)) {
      Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
      unsigned Code = getICmpCode(LHS) | getICmpCode(RHS);
      bool isSigned = LHS->isSigned() || RHS->isSigned();
      return getNewICmpValue(isSigned, Code, Op0, Op1, Builder);
    }
  }

  // handle (roughly):
  // (icmp ne (A & B), C) | (icmp ne (A & D), E)
  if (Value *V = foldLogOpOfMaskedICmps(LHS, RHS, false, Builder))
    return V;

  Value *Val = LHS->getOperand(0), *Val2 = RHS->getOperand(0);
  if (LHS->hasOneUse() || RHS->hasOneUse()) {
    // (icmp eq B, 0) | (icmp ult A, B) -> (icmp ule A, B-1)
    // (icmp eq B, 0) | (icmp ugt B, A) -> (icmp ule A, B-1)
    Value *A = nullptr, *B = nullptr;
    if (LHSCC == ICmpInst::ICMP_EQ && LHSCst && LHSCst->isZero()) {
      B = Val;
      if (RHSCC == ICmpInst::ICMP_ULT && Val == RHS->getOperand(1))
        A = Val2;
      else if (RHSCC == ICmpInst::ICMP_UGT && Val == Val2)
        A = RHS->getOperand(1);
    }
    // (icmp ult A, B) | (icmp eq B, 0) -> (icmp ule A, B-1)
    // (icmp ugt B, A) | (icmp eq B, 0) -> (icmp ule A, B-1)
    else if (RHSCC == ICmpInst::ICMP_EQ && RHSCst && RHSCst->isZero()) {
      B = Val2;
      if (LHSCC == ICmpInst::ICMP_ULT && Val2 == LHS->getOperand(1))
        A = Val;
      else if (LHSCC == ICmpInst::ICMP_UGT && Val2 == Val)
        A = LHS->getOperand(1);
    }
    if (A && B)
      return Builder->CreateICmp(
          ICmpInst::ICMP_UGE,
          Builder->CreateAdd(B, ConstantInt::getSigned(B->getType(), -1)), A);
  }

  // This only handles icmp of constants: (icmp1 A, C1) | (icmp2 B, C2).
  if (!LHSCst || !RHSCst) return nullptr;

  if (LHSCst == RHSCst && LHSCC == RHSCC) {
    // (icmp ne A, 0) | (icmp ne B, 0) --> (icmp ne (A|B), 0)
    if (LHSCC == ICmpInst::ICMP_NE && LHSCst->isZero()) {
      Value *NewOr = Builder->CreateOr(Val, Val2);
      return Builder->CreateICmp(LHSCC, NewOr, LHSCst);
    }
  }

  // (icmp ult (X + CA), C1) | (icmp eq X, C2) -> (icmp ule (X + CA), C1)
  //   iff C2 + CA == C1.
  if (LHSCC == ICmpInst::ICMP_ULT && RHSCC == ICmpInst::ICMP_EQ) {
    ConstantInt *AddCst;
    if (match(Val, m_Add(m_Specific(Val2), m_ConstantInt(AddCst))))
      if (RHSCst->getValue() + AddCst->getValue() == LHSCst->getValue())
        return Builder->CreateICmpULE(Val, LHSCst);
  }

  // From here on, we only handle:
  //    (icmp1 A, C1) | (icmp2 A, C2) --> something simpler.
  if (Val != Val2) return nullptr;

  // ICMP_[US][GL]E X, CST is folded to ICMP_[US][GL]T elsewhere.
  if (LHSCC == ICmpInst::ICMP_UGE || LHSCC == ICmpInst::ICMP_ULE ||
      RHSCC == ICmpInst::ICMP_UGE || RHSCC == ICmpInst::ICMP_ULE ||
      LHSCC == ICmpInst::ICMP_SGE || LHSCC == ICmpInst::ICMP_SLE ||
      RHSCC == ICmpInst::ICMP_SGE || RHSCC == ICmpInst::ICMP_SLE)
    return nullptr;

  // We can't fold (ugt x, C) | (sgt x, C2).
  if (!PredicatesFoldable(LHSCC, RHSCC))
    return nullptr;

  // Ensure that the larger constant is on the RHS.
  bool ShouldSwap;
  if (CmpInst::isSigned(LHSCC) ||
      (ICmpInst::isEquality(LHSCC) &&
       CmpInst::isSigned(RHSCC)))
    ShouldSwap = LHSCst->getValue().sgt(RHSCst->getValue());
  else
    ShouldSwap = LHSCst->getValue().ugt(RHSCst->getValue());

  if (ShouldSwap) {
    std::swap(LHS, RHS);
    std::swap(LHSCst, RHSCst);
    std::swap(LHSCC, RHSCC);
  }

  // At this point, we know we have two icmp instructions
  // comparing a value against two constants and or'ing the result
  // together.  Because of the above check, we know that we only have
  // ICMP_EQ, ICMP_NE, ICMP_LT, and ICMP_GT here. We also know (from the
  // icmp folding check above), that the two constants are not
  // equal.
  assert(LHSCst != RHSCst && "Compares not folded above?");

  switch (LHSCC) {
  default: llvm_unreachable("Unknown integer condition code!");
  case ICmpInst::ICMP_EQ:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:
      if (LHS->getOperand(0) == RHS->getOperand(0)) {
        // if LHSCst and RHSCst differ only by one bit:
        // (A == C1 || A == C2) -> (A & ~(C1 ^ C2)) == C1
        assert(LHSCst->getValue().ule(LHSCst->getValue()));

        APInt Xor = LHSCst->getValue() ^ RHSCst->getValue();
        if (Xor.isPowerOf2()) {
          Value *NegCst = Builder->getInt(~Xor);
          Value *And = Builder->CreateAnd(LHS->getOperand(0), NegCst);
          return Builder->CreateICmp(ICmpInst::ICMP_EQ, And, LHSCst);
        }
      }

      if (LHSCst == SubOne(RHSCst)) {
        // (X == 13 | X == 14) -> X-13 <u 2
        Constant *AddCST = ConstantExpr::getNeg(LHSCst);
        Value *Add = Builder->CreateAdd(Val, AddCST, Val->getName()+".off");
        AddCST = ConstantExpr::getSub(AddOne(RHSCst), LHSCst);
        return Builder->CreateICmpULT(Add, AddCST);
      }

      break;                         // (X == 13 | X == 15) -> no change
    case ICmpInst::ICMP_UGT:         // (X == 13 | X u> 14) -> no change
    case ICmpInst::ICMP_SGT:         // (X == 13 | X s> 14) -> no change
      break;
    case ICmpInst::ICMP_NE:          // (X == 13 | X != 15) -> X != 15
    case ICmpInst::ICMP_ULT:         // (X == 13 | X u< 15) -> X u< 15
    case ICmpInst::ICMP_SLT:         // (X == 13 | X s< 15) -> X s< 15
      return RHS;
    }
    break;
  case ICmpInst::ICMP_NE:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:          // (X != 13 | X == 15) -> X != 13
    case ICmpInst::ICMP_UGT:         // (X != 13 | X u> 15) -> X != 13
    case ICmpInst::ICMP_SGT:         // (X != 13 | X s> 15) -> X != 13
      return LHS;
    case ICmpInst::ICMP_NE:          // (X != 13 | X != 15) -> true
    case ICmpInst::ICMP_ULT:         // (X != 13 | X u< 15) -> true
    case ICmpInst::ICMP_SLT:         // (X != 13 | X s< 15) -> true
      return Builder->getTrue();
    }
  case ICmpInst::ICMP_ULT:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:         // (X u< 13 | X == 14) -> no change
      break;
    case ICmpInst::ICMP_UGT:        // (X u< 13 | X u> 15) -> (X-13) u> 2
      // If RHSCst is [us]MAXINT, it is always false.  Not handling
      // this can cause overflow.
      if (RHSCst->isMaxValue(false))
        return LHS;
      return InsertRangeTest(Val, LHSCst, AddOne(RHSCst), false, false);
    case ICmpInst::ICMP_SGT:        // (X u< 13 | X s> 15) -> no change
      break;
    case ICmpInst::ICMP_NE:         // (X u< 13 | X != 15) -> X != 15
    case ICmpInst::ICMP_ULT:        // (X u< 13 | X u< 15) -> X u< 15
      return RHS;
    case ICmpInst::ICMP_SLT:        // (X u< 13 | X s< 15) -> no change
      break;
    }
    break;
  case ICmpInst::ICMP_SLT:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:         // (X s< 13 | X == 14) -> no change
      break;
    case ICmpInst::ICMP_SGT:        // (X s< 13 | X s> 15) -> (X-13) s> 2
      // If RHSCst is [us]MAXINT, it is always false.  Not handling
      // this can cause overflow.
      if (RHSCst->isMaxValue(true))
        return LHS;
      return InsertRangeTest(Val, LHSCst, AddOne(RHSCst), true, false);
    case ICmpInst::ICMP_UGT:        // (X s< 13 | X u> 15) -> no change
      break;
    case ICmpInst::ICMP_NE:         // (X s< 13 | X != 15) -> X != 15
    case ICmpInst::ICMP_SLT:        // (X s< 13 | X s< 15) -> X s< 15
      return RHS;
    case ICmpInst::ICMP_ULT:        // (X s< 13 | X u< 15) -> no change
      break;
    }
    break;
  case ICmpInst::ICMP_UGT:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:         // (X u> 13 | X == 15) -> X u> 13
    case ICmpInst::ICMP_UGT:        // (X u> 13 | X u> 15) -> X u> 13
      return LHS;
    case ICmpInst::ICMP_SGT:        // (X u> 13 | X s> 15) -> no change
      break;
    case ICmpInst::ICMP_NE:         // (X u> 13 | X != 15) -> true
    case ICmpInst::ICMP_ULT:        // (X u> 13 | X u< 15) -> true
      return Builder->getTrue();
    case ICmpInst::ICMP_SLT:        // (X u> 13 | X s< 15) -> no change
      break;
    }
    break;
  case ICmpInst::ICMP_SGT:
    switch (RHSCC) {
    default: llvm_unreachable("Unknown integer condition code!");
    case ICmpInst::ICMP_EQ:         // (X s> 13 | X == 15) -> X > 13
    case ICmpInst::ICMP_SGT:        // (X s> 13 | X s> 15) -> X > 13
      return LHS;
    case ICmpInst::ICMP_UGT:        // (X s> 13 | X u> 15) -> no change
      break;
    case ICmpInst::ICMP_NE:         // (X s> 13 | X != 15) -> true
    case ICmpInst::ICMP_SLT:        // (X s> 13 | X s< 15) -> true
      return Builder->getTrue();
    case ICmpInst::ICMP_ULT:        // (X s> 13 | X u< 15) -> no change
      break;
    }
    break;
  }
  return nullptr;
}

Instruction * InstCombiner::FoldOrWithConstants	(	BinaryOperator &	I,
		Value *	Op,
		Value *	A,
		Value *	B,
		Value *	C
	)

FoldOrWithConstants - This helper function folds:

((A | B) & C1) | (B & C2)

into:

(A & C1) | B

when the XOR of the two constants is "all ones" (-1).

Definition at line 1889 of file InstCombineAndOrXor.cpp.

                                                                             {
  ConstantInt *CI1 = dyn_cast<ConstantInt>(C);
  if (!CI1) return nullptr;

  Value *V1 = nullptr;
  ConstantInt *CI2 = nullptr;
  if (!match(Op, m_And(m_Value(V1), m_ConstantInt(CI2)))) return nullptr;

  APInt Xor = CI1->getValue() ^ CI2->getValue();
  if (!Xor.isAllOnesValue()) return nullptr;

  if (V1 == A || V1 == B) {
    Value *NewOp = Builder->CreateAnd((V1 == A) ? B : A, CI1);
    return BinaryOperator::CreateOr(NewOp, V1);
  }

  return nullptr;
}

Instruction * InstCombiner::FoldPHIArgBinOpIntoPHI ( PHINode &  PN )

private

FoldPHIArgBinOpIntoPHI - If we have something like phi [add (a,b), add(a,c)] and if a/b/c and the add's all have a single use, turn this into a phi and a single binop.

Definition at line 26 of file InstCombinePHI.cpp.

                                                             {
  Instruction *FirstInst = cast<Instruction>(PN.getIncomingValue(0));
  assert(isa<BinaryOperator>(FirstInst) || isa<CmpInst>(FirstInst));
  unsigned Opc = FirstInst->getOpcode();
  Value *LHSVal = FirstInst->getOperand(0);
  Value *RHSVal = FirstInst->getOperand(1);

  Type *LHSType = LHSVal->getType();
  Type *RHSType = RHSVal->getType();

  bool isNUW = false, isNSW = false, isExact = false;
  if (OverflowingBinaryOperator *BO =
        dyn_cast<OverflowingBinaryOperator>(FirstInst)) {
    isNUW = BO->hasNoUnsignedWrap();
    isNSW = BO->hasNoSignedWrap();
  } else if (PossiblyExactOperator *PEO =
               dyn_cast<PossiblyExactOperator>(FirstInst))
    isExact = PEO->isExact();

  // Scan to see if all operands are the same opcode, and all have one use.
  for (unsigned i = 1; i != PN.getNumIncomingValues(); ++i) {
    Instruction *I = dyn_cast<Instruction>(PN.getIncomingValue(i));
    if (!I || I->getOpcode() != Opc || !I->hasOneUse() ||
        // Verify type of the LHS matches so we don't fold cmp's of different
        // types.
        I->getOperand(0)->getType() != LHSType ||
        I->getOperand(1)->getType() != RHSType)
      return nullptr;

    // If they are CmpInst instructions, check their predicates
    if (CmpInst *CI = dyn_cast<CmpInst>(I))
      if (CI->getPredicate() != cast<CmpInst>(FirstInst)->getPredicate())
        return nullptr;

    if (isNUW)
      isNUW = cast<OverflowingBinaryOperator>(I)->hasNoUnsignedWrap();
    if (isNSW)
      isNSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
    if (isExact)
      isExact = cast<PossiblyExactOperator>(I)->isExact();

    // Keep track of which operand needs a phi node.
    if (I->getOperand(0) != LHSVal) LHSVal = nullptr;
    if (I->getOperand(1) != RHSVal) RHSVal = nullptr;
  }

  // If both LHS and RHS would need a PHI, don't do this transformation,
  // because it would increase the number of PHIs entering the block,
  // which leads to higher register pressure. This is especially
  // bad when the PHIs are in the header of a loop.
  if (!LHSVal && !RHSVal)
    return nullptr;

  // Otherwise, this is safe to transform!

  Value *InLHS = FirstInst->getOperand(0);
  Value *InRHS = FirstInst->getOperand(1);
  PHINode *NewLHS = nullptr, *NewRHS = nullptr;
  if (!LHSVal) {
    NewLHS = PHINode::Create(LHSType, PN.getNumIncomingValues(),
                             FirstInst->getOperand(0)->getName() + ".pn");
    NewLHS->addIncoming(InLHS, PN.getIncomingBlock(0));
    InsertNewInstBefore(NewLHS, PN);
    LHSVal = NewLHS;
  }

  if (!RHSVal) {
    NewRHS = PHINode::Create(RHSType, PN.getNumIncomingValues(),
                             FirstInst->getOperand(1)->getName() + ".pn");
    NewRHS->addIncoming(InRHS, PN.getIncomingBlock(0));
    InsertNewInstBefore(NewRHS, PN);
    RHSVal = NewRHS;
  }

  // Add all operands to the new PHIs.
  if (NewLHS || NewRHS) {
    for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) {
      Instruction *InInst = cast<Instruction>(PN.getIncomingValue(i));
      if (NewLHS) {
        Value *NewInLHS = InInst->getOperand(0);
        NewLHS->addIncoming(NewInLHS, PN.getIncomingBlock(i));
      }
      if (NewRHS) {
        Value *NewInRHS = InInst->getOperand(1);
        NewRHS->addIncoming(NewInRHS, PN.getIncomingBlock(i));
      }
    }
  }

  if (CmpInst *CIOp = dyn_cast<CmpInst>(FirstInst)) {
    CmpInst *NewCI = CmpInst::Create(CIOp->getOpcode(), CIOp->getPredicate(),
                                     LHSVal, RHSVal);
    NewCI->setDebugLoc(FirstInst->getDebugLoc());
    return NewCI;
  }

  BinaryOperator *BinOp = cast<BinaryOperator>(FirstInst);
  BinaryOperator *NewBinOp =
    BinaryOperator::Create(BinOp->getOpcode(), LHSVal, RHSVal);
  if (isNUW) NewBinOp->setHasNoUnsignedWrap();
  if (isNSW) NewBinOp->setHasNoSignedWrap();
  if (isExact) NewBinOp->setIsExact();
  NewBinOp->setDebugLoc(FirstInst->getDebugLoc());
  return NewBinOp;
}

Instruction * InstCombiner::FoldPHIArgGEPIntoPHI ( PHINode &  PN )

private

Definition at line 132 of file InstCombinePHI.cpp.

                                                           {
  GetElementPtrInst *FirstInst =cast<GetElementPtrInst>(PN.getIncomingValue(0));

  SmallVector<Value*, 16> FixedOperands(FirstInst->op_begin(),
                                        FirstInst->op_end());
  // This is true if all GEP bases are allocas and if all indices into them are
  // constants.
  bool AllBasePointersAreAllocas = true;

  // We don't want to replace this phi if the replacement would require
  // more than one phi, which leads to higher register pressure. This is
  // especially bad when the PHIs are in the header of a loop.
  bool NeededPhi = false;

  bool AllInBounds = true;

  // Scan to see if all operands are the same opcode, and all have one use.
  for (unsigned i = 1; i != PN.getNumIncomingValues(); ++i) {
    GetElementPtrInst *GEP= dyn_cast<GetElementPtrInst>(PN.getIncomingValue(i));
    if (!GEP || !GEP->hasOneUse() || GEP->getType() != FirstInst->getType() ||
      GEP->getNumOperands() != FirstInst->getNumOperands())
      return nullptr;

    AllInBounds &= GEP->isInBounds();

    // Keep track of whether or not all GEPs are of alloca pointers.
    if (AllBasePointersAreAllocas &&
        (!isa<AllocaInst>(GEP->getOperand(0)) ||
         !GEP->hasAllConstantIndices()))
      AllBasePointersAreAllocas = false;

    // Compare the operand lists.
    for (unsigned op = 0, e = FirstInst->getNumOperands(); op != e; ++op) {
      if (FirstInst->getOperand(op) == GEP->getOperand(op))
        continue;

      // Don't merge two GEPs when two operands differ (introducing phi nodes)
      // if one of the PHIs has a constant for the index.  The index may be
      // substantially cheaper to compute for the constants, so making it a
      // variable index could pessimize the path.  This also handles the case
      // for struct indices, which must always be constant.
      if (isa<ConstantInt>(FirstInst->getOperand(op)) ||
          isa<ConstantInt>(GEP->getOperand(op)))
        return nullptr;

      if (FirstInst->getOperand(op)->getType() !=GEP->getOperand(op)->getType())
        return nullptr;

      // If we already needed a PHI for an earlier operand, and another operand
      // also requires a PHI, we'd be introducing more PHIs than we're
      // eliminating, which increases register pressure on entry to the PHI's
      // block.
      if (NeededPhi)
        return nullptr;

      FixedOperands[op] = nullptr;  // Needs a PHI.
      NeededPhi = true;
    }
  }

  // If all of the base pointers of the PHI'd GEPs are from allocas, don't
  // bother doing this transformation.  At best, this will just save a bit of
  // offset calculation, but all the predecessors will have to materialize the
  // stack address into a register anyway.  We'd actually rather *clone* the
  // load up into the predecessors so that we have a load of a gep of an alloca,
  // which can usually all be folded into the load.
  if (AllBasePointersAreAllocas)
    return nullptr;

  // Otherwise, this is safe to transform.  Insert PHI nodes for each operand
  // that is variable.
  SmallVector<PHINode*, 16> OperandPhis(FixedOperands.size());

  bool HasAnyPHIs = false;
  for (unsigned i = 0, e = FixedOperands.size(); i != e; ++i) {
    if (FixedOperands[i]) continue;  // operand doesn't need a phi.
    Value *FirstOp = FirstInst->getOperand(i);
    PHINode *NewPN = PHINode::Create(FirstOp->getType(), e,
                                     FirstOp->getName()+".pn");
    InsertNewInstBefore(NewPN, PN);

    NewPN->addIncoming(FirstOp, PN.getIncomingBlock(0));
    OperandPhis[i] = NewPN;
    FixedOperands[i] = NewPN;
    HasAnyPHIs = true;
  }


  // Add all operands to the new PHIs.
  if (HasAnyPHIs) {
    for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) {
      GetElementPtrInst *InGEP =cast<GetElementPtrInst>(PN.getIncomingValue(i));
      BasicBlock *InBB = PN.getIncomingBlock(i);

      for (unsigned op = 0, e = OperandPhis.size(); op != e; ++op)
        if (PHINode *OpPhi = OperandPhis[op])
          OpPhi->addIncoming(InGEP->getOperand(op), InBB);
    }
  }

  Value *Base = FixedOperands[0];
  GetElementPtrInst *NewGEP =
    GetElementPtrInst::Create(Base, makeArrayRef(FixedOperands).slice(1));
  if (AllInBounds) NewGEP->setIsInBounds();
  NewGEP->setDebugLoc(FirstInst->getDebugLoc());
  return NewGEP;
}

Instruction * InstCombiner::FoldPHIArgLoadIntoPHI ( PHINode &  PN )

private

Definition at line 287 of file InstCombinePHI.cpp.

                                                            {
  LoadInst *FirstLI = cast<LoadInst>(PN.getIncomingValue(0));

  // FIXME: This is overconservative; this transform is allowed in some cases
  // for atomic operations.
  if (FirstLI->isAtomic())
    return nullptr;

  // When processing loads, we need to propagate two bits of information to the
  // sunk load: whether it is volatile, and what its alignment is.  We currently
  // don't sink loads when some have their alignment specified and some don't.
  // visitLoadInst will propagate an alignment onto the load when TD is around,
  // and if TD isn't around, we can't handle the mixed case.
  bool isVolatile = FirstLI->isVolatile();
  unsigned LoadAlignment = FirstLI->getAlignment();
  unsigned LoadAddrSpace = FirstLI->getPointerAddressSpace();

  // We can't sink the load if the loaded value could be modified between the
  // load and the PHI.
  if (FirstLI->getParent() != PN.getIncomingBlock(0) ||
      !isSafeAndProfitableToSinkLoad(FirstLI))
    return nullptr;

  // If the PHI is of volatile loads and the load block has multiple
  // successors, sinking it would remove a load of the volatile value from
  // the path through the other successor.
  if (isVolatile &&
      FirstLI->getParent()->getTerminator()->getNumSuccessors() != 1)
    return nullptr;

  // Check to see if all arguments are the same operation.
  for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) {
    LoadInst *LI = dyn_cast<LoadInst>(PN.getIncomingValue(i));
    if (!LI || !LI->hasOneUse())
      return nullptr;

    // We can't sink the load if the loaded value could be modified between
    // the load and the PHI.
    if (LI->isVolatile() != isVolatile ||
        LI->getParent() != PN.getIncomingBlock(i) ||
        LI->getPointerAddressSpace() != LoadAddrSpace ||
        !isSafeAndProfitableToSinkLoad(LI))
      return nullptr;

    // If some of the loads have an alignment specified but not all of them,
    // we can't do the transformation.
    if ((LoadAlignment != 0) != (LI->getAlignment() != 0))
      return nullptr;

    LoadAlignment = std::min(LoadAlignment, LI->getAlignment());

    // If the PHI is of volatile loads and the load block has multiple
    // successors, sinking it would remove a load of the volatile value from
    // the path through the other successor.
    if (isVolatile &&
        LI->getParent()->getTerminator()->getNumSuccessors() != 1)
      return nullptr;
  }

  // Okay, they are all the same operation.  Create a new PHI node of the
  // correct type, and PHI together all of the LHS's of the instructions.
  PHINode *NewPN = PHINode::Create(FirstLI->getOperand(0)->getType(),
                                   PN.getNumIncomingValues(),
                                   PN.getName()+".in");

  Value *InVal = FirstLI->getOperand(0);
  NewPN->addIncoming(InVal, PN.getIncomingBlock(0));

  // Add all operands to the new PHI.
  for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) {
    Value *NewInVal = cast<LoadInst>(PN.getIncomingValue(i))->getOperand(0);
    if (NewInVal != InVal)
      InVal = nullptr;
    NewPN->addIncoming(NewInVal, PN.getIncomingBlock(i));
  }

  Value *PhiVal;
  if (InVal) {
    // The new PHI unions all of the same values together.  This is really
    // common, so we handle it intelligently here for compile-time speed.
    PhiVal = InVal;
    delete NewPN;
  } else {
    InsertNewInstBefore(NewPN, PN);
    PhiVal = NewPN;
  }

  // If this was a volatile load that we are merging, make sure to loop through
  // and mark all the input loads as non-volatile.  If we don't do this, we will
  // insert a new volatile load and the old ones will not be deletable.
  if (isVolatile)
    for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i)
      cast<LoadInst>(PN.getIncomingValue(i))->setVolatile(false);

  LoadInst *NewLI = new LoadInst(PhiVal, "", isVolatile, LoadAlignment);
  NewLI->setDebugLoc(FirstLI->getDebugLoc());
  return NewLI;
}

Instruction * InstCombiner::FoldPHIArgOpIntoPHI ( PHINode &  PN )

private

FoldPHIArgOpIntoPHI - If all operands to a PHI node are the same "unary" operator and they all are only used by the PHI, PHI together their inputs, and do the operation once, to the result of the PHI.

Definition at line 391 of file InstCombinePHI.cpp.

                                                          {
  Instruction *FirstInst = cast<Instruction>(PN.getIncomingValue(0));

  if (isa<GetElementPtrInst>(FirstInst))
    return FoldPHIArgGEPIntoPHI(PN);
  if (isa<LoadInst>(FirstInst))
    return FoldPHIArgLoadIntoPHI(PN);

  // Scan the instruction, looking for input operations that can be folded away.
  // If all input operands to the phi are the same instruction (e.g. a cast from
  // the same type or "+42") we can pull the operation through the PHI, reducing
  // code size and simplifying code.
  Constant *ConstantOp = nullptr;
  Type *CastSrcTy = nullptr;
  bool isNUW = false, isNSW = false, isExact = false;

  if (isa<CastInst>(FirstInst)) {
    CastSrcTy = FirstInst->getOperand(0)->getType();

    // Be careful about transforming integer PHIs.  We don't want to pessimize
    // the code by turning an i32 into an i1293.
    if (PN.getType()->isIntegerTy() && CastSrcTy->isIntegerTy()) {
      if (!ShouldChangeType(PN.getType(), CastSrcTy))
        return nullptr;
    }
  } else if (isa<BinaryOperator>(FirstInst) || isa<CmpInst>(FirstInst)) {
    // Can fold binop, compare or shift here if the RHS is a constant,
    // otherwise call FoldPHIArgBinOpIntoPHI.
    ConstantOp = dyn_cast<Constant>(FirstInst->getOperand(1));
    if (!ConstantOp)
      return FoldPHIArgBinOpIntoPHI(PN);

    if (OverflowingBinaryOperator *BO =
        dyn_cast<OverflowingBinaryOperator>(FirstInst)) {
      isNUW = BO->hasNoUnsignedWrap();
      isNSW = BO->hasNoSignedWrap();
    } else if (PossiblyExactOperator *PEO =
               dyn_cast<PossiblyExactOperator>(FirstInst))
      isExact = PEO->isExact();
  } else {
    return nullptr;  // Cannot fold this operation.
  }

  // Check to see if all arguments are the same operation.
  for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) {
    Instruction *I = dyn_cast<Instruction>(PN.getIncomingValue(i));
    if (!I || !I->hasOneUse() || !I->isSameOperationAs(FirstInst))
      return nullptr;
    if (CastSrcTy) {
      if (I->getOperand(0)->getType() != CastSrcTy)
        return nullptr;  // Cast operation must match.
    } else if (I->getOperand(1) != ConstantOp) {
      return nullptr;
    }

    if (isNUW)
      isNUW = cast<OverflowingBinaryOperator>(I)->hasNoUnsignedWrap();
    if (isNSW)
      isNSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
    if (isExact)
      isExact = cast<PossiblyExactOperator>(I)->isExact();
  }

  // Okay, they are all the same operation.  Create a new PHI node of the
  // correct type, and PHI together all of the LHS's of the instructions.
  PHINode *NewPN = PHINode::Create(FirstInst->getOperand(0)->getType(),
                                   PN.getNumIncomingValues(),
                                   PN.getName()+".in");

  Value *InVal = FirstInst->getOperand(0);
  NewPN->addIncoming(InVal, PN.getIncomingBlock(0));

  // Add all operands to the new PHI.
  for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) {
    Value *NewInVal = cast<Instruction>(PN.getIncomingValue(i))->getOperand(0);
    if (NewInVal != InVal)
      InVal = nullptr;
    NewPN->addIncoming(NewInVal, PN.getIncomingBlock(i));
  }

  Value *PhiVal;
  if (InVal) {
    // The new PHI unions all of the same values together.  This is really
    // common, so we handle it intelligently here for compile-time speed.
    PhiVal = InVal;
    delete NewPN;
  } else {
    InsertNewInstBefore(NewPN, PN);
    PhiVal = NewPN;
  }

  // Insert and return the new operation.
  if (CastInst *FirstCI = dyn_cast<CastInst>(FirstInst)) {
    CastInst *NewCI = CastInst::Create(FirstCI->getOpcode(), PhiVal,
                                       PN.getType());
    NewCI->setDebugLoc(FirstInst->getDebugLoc());
    return NewCI;
  }

  if (BinaryOperator *BinOp = dyn_cast<BinaryOperator>(FirstInst)) {
    BinOp = BinaryOperator::Create(BinOp->getOpcode(), PhiVal, ConstantOp);
    if (isNUW) BinOp->setHasNoUnsignedWrap();
    if (isNSW) BinOp->setHasNoSignedWrap();
    if (isExact) BinOp->setIsExact();
    BinOp->setDebugLoc(FirstInst->getDebugLoc());
    return BinOp;
  }

  CmpInst *CIOp = cast<CmpInst>(FirstInst);
  CmpInst *NewCI = CmpInst::Create(CIOp->getOpcode(), CIOp->getPredicate(),
                                   PhiVal, ConstantOp);
  NewCI->setDebugLoc(FirstInst->getDebugLoc());
  return NewCI;
}

Instruction * InstCombiner::FoldSelectIntoOp	(	SelectInst &	SI,
		Value *	TrueVal,
		Value *	FalseVal
	)

FoldSelectIntoOp - Try fold the select into one of the operands to facilitate further optimization.

Definition at line 235 of file InstCombineSelect.cpp.

                                                             {
  // See the comment above GetSelectFoldableOperands for a description of the
  // transformation we are doing here.
  if (Instruction *TVI = dyn_cast<Instruction>(TrueVal)) {
    if (TVI->hasOneUse() && TVI->getNumOperands() == 2 &&
        !isa<Constant>(FalseVal)) {
      if (unsigned SFO = GetSelectFoldableOperands(TVI)) {
        unsigned OpToFold = 0;
        if ((SFO & 1) && FalseVal == TVI->getOperand(0)) {
          OpToFold = 1;
        } else if ((SFO & 2) && FalseVal == TVI->getOperand(1)) {
          OpToFold = 2;
        }

        if (OpToFold) {
          Constant *C = GetSelectFoldableConstant(TVI);
          Value *OOp = TVI->getOperand(2-OpToFold);
          // Avoid creating select between 2 constants unless it's selecting
          // between 0, 1 and -1.
          if (!isa<Constant>(OOp) || isSelect01(C, cast<Constant>(OOp))) {
            Value *NewSel = Builder->CreateSelect(SI.getCondition(), OOp, C);
            NewSel->takeName(TVI);
            BinaryOperator *TVI_BO = cast<BinaryOperator>(TVI);
            BinaryOperator *BO = BinaryOperator::Create(TVI_BO->getOpcode(),
                                                        FalseVal, NewSel);
            if (isa<PossiblyExactOperator>(BO))
              BO->setIsExact(TVI_BO->isExact());
            if (isa<OverflowingBinaryOperator>(BO)) {
              BO->setHasNoUnsignedWrap(TVI_BO->hasNoUnsignedWrap());
              BO->setHasNoSignedWrap(TVI_BO->hasNoSignedWrap());
            }
            return BO;
          }
        }
      }
    }
  }

  if (Instruction *FVI = dyn_cast<Instruction>(FalseVal)) {
    if (FVI->hasOneUse() && FVI->getNumOperands() == 2 &&
        !isa<Constant>(TrueVal)) {
      if (unsigned SFO = GetSelectFoldableOperands(FVI)) {
        unsigned OpToFold = 0;
        if ((SFO & 1) && TrueVal == FVI->getOperand(0)) {
          OpToFold = 1;
        } else if ((SFO & 2) && TrueVal == FVI->getOperand(1)) {
          OpToFold = 2;
        }

        if (OpToFold) {
          Constant *C = GetSelectFoldableConstant(FVI);
          Value *OOp = FVI->getOperand(2-OpToFold);
          // Avoid creating select between 2 constants unless it's selecting
          // between 0, 1 and -1.
          if (!isa<Constant>(OOp) || isSelect01(C, cast<Constant>(OOp))) {
            Value *NewSel = Builder->CreateSelect(SI.getCondition(), C, OOp);
            NewSel->takeName(FVI);
            BinaryOperator *FVI_BO = cast<BinaryOperator>(FVI);
            BinaryOperator *BO = BinaryOperator::Create(FVI_BO->getOpcode(),
                                                        TrueVal, NewSel);
            if (isa<PossiblyExactOperator>(BO))
              BO->setIsExact(FVI_BO->isExact());
            if (isa<OverflowingBinaryOperator>(BO)) {
              BO->setHasNoUnsignedWrap(FVI_BO->hasNoUnsignedWrap());
              BO->setHasNoSignedWrap(FVI_BO->hasNoSignedWrap());
            }
            return BO;
          }
        }
      }
    }
  }

  return nullptr;
}

Instruction * InstCombiner::FoldSelectOpOp	(	SelectInst &	SI,
		Instruction *	TI,
		Instruction *	FI
	)

FoldSelectOpOp - Here we have (select c, TI, FI), and we know that TI and FI have the same opcode and only one use each. Try to simplify this.

Definition at line 148 of file InstCombineSelect.cpp.

                                                           {
  if (TI->getNumOperands() == 1) {
    // If this is a non-volatile load or a cast from the same type,
    // merge.
    if (TI->isCast()) {
      Type *FIOpndTy = FI->getOperand(0)->getType();
      if (TI->getOperand(0)->getType() != FIOpndTy)
        return nullptr;
      // The select condition may be a vector. We may only change the operand
      // type if the vector width remains the same (and matches the condition).
      Type *CondTy = SI.getCondition()->getType();
      if (CondTy->isVectorTy() && (!FIOpndTy->isVectorTy() ||
          CondTy->getVectorNumElements() != FIOpndTy->getVectorNumElements()))
        return nullptr;
    } else {
      return nullptr;  // unknown unary op.
    }

    // Fold this by inserting a select from the input values.
    Value *NewSI = Builder->CreateSelect(SI.getCondition(), TI->getOperand(0),
                                         FI->getOperand(0), SI.getName()+".v");
    return CastInst::Create(Instruction::CastOps(TI->getOpcode()), NewSI,
                            TI->getType());
  }

  // Only handle binary operators here.
  if (!isa<BinaryOperator>(TI))
    return nullptr;

  // Figure out if the operations have any operands in common.
  Value *MatchOp, *OtherOpT, *OtherOpF;
  bool MatchIsOpZero;
  if (TI->getOperand(0) == FI->getOperand(0)) {
    MatchOp  = TI->getOperand(0);
    OtherOpT = TI->getOperand(1);
    OtherOpF = FI->getOperand(1);
    MatchIsOpZero = true;
  } else if (TI->getOperand(1) == FI->getOperand(1)) {
    MatchOp  = TI->getOperand(1);
    OtherOpT = TI->getOperand(0);
    OtherOpF = FI->getOperand(0);
    MatchIsOpZero = false;
  } else if (!TI->isCommutative()) {
    return nullptr;
  } else if (TI->getOperand(0) == FI->getOperand(1)) {
    MatchOp  = TI->getOperand(0);
    OtherOpT = TI->getOperand(1);
    OtherOpF = FI->getOperand(0);
    MatchIsOpZero = true;
  } else if (TI->getOperand(1) == FI->getOperand(0)) {
    MatchOp  = TI->getOperand(1);
    OtherOpT = TI->getOperand(0);
    OtherOpF = FI->getOperand(1);
    MatchIsOpZero = true;
  } else {
    return nullptr;
  }

  // If we reach here, they do have operations in common.
  Value *NewSI = Builder->CreateSelect(SI.getCondition(), OtherOpT,
                                       OtherOpF, SI.getName()+".v");

  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(TI)) {
    if (MatchIsOpZero)
      return BinaryOperator::Create(BO->getOpcode(), MatchOp, NewSI);
    else
      return BinaryOperator::Create(BO->getOpcode(), NewSI, MatchOp);
  }
  llvm_unreachable("Shouldn't get here");
}

Instruction * InstCombiner::FoldShiftByConstant	(	Value *	Op0,
		Constant *	Op1,
		BinaryOperator &	I
	)

Definition at line 314 of file InstCombineShifts.cpp.

                                                                  {
  bool isLeftShift = I.getOpcode() == Instruction::Shl;

  ConstantInt *COp1 = nullptr;
  if (ConstantDataVector *CV = dyn_cast<ConstantDataVector>(Op1))
    COp1 = dyn_cast_or_null<ConstantInt>(CV->getSplatValue());
  else if (ConstantVector *CV = dyn_cast<ConstantVector>(Op1))
    COp1 = dyn_cast_or_null<ConstantInt>(CV->getSplatValue());
  else
    COp1 = dyn_cast<ConstantInt>(Op1);

  if (!COp1)
    return nullptr;

  // See if we can propagate this shift into the input, this covers the trivial
  // cast of lshr(shl(x,c1),c2) as well as other more complex cases.
  if (I.getOpcode() != Instruction::AShr &&
      CanEvaluateShifted(Op0, COp1->getZExtValue(), isLeftShift, *this)) {
    DEBUG(dbgs() << "ICE: GetShiftedValue propagating shift through expression"
              " to eliminate shift:\n  IN: " << *Op0 << "\n  SH: " << I <<"\n");

    return ReplaceInstUsesWith(I,
                 GetShiftedValue(Op0, COp1->getZExtValue(), isLeftShift, *this));
  }

  // See if we can simplify any instructions used by the instruction whose sole
  // purpose is to compute bits we don't care about.
  uint32_t TypeBits = Op0->getType()->getScalarSizeInBits();

  assert(!COp1->uge(TypeBits) &&
         "Shift over the type width should have been removed already");

  // ((X*C1) << C2) == (X * (C1 << C2))
  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Op0))
    if (BO->getOpcode() == Instruction::Mul && isLeftShift)
      if (Constant *BOOp = dyn_cast<Constant>(BO->getOperand(1)))
        return BinaryOperator::CreateMul(BO->getOperand(0),
                                        ConstantExpr::getShl(BOOp, Op1));

  // Try to fold constant and into select arguments.
  if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
    if (Instruction *R = FoldOpIntoSelect(I, SI))
      return R;
  if (isa<PHINode>(Op0))
    if (Instruction *NV = FoldOpIntoPhi(I))
      return NV;

  // Fold shift2(trunc(shift1(x,c1)), c2) -> trunc(shift2(shift1(x,c1),c2))
  if (TruncInst *TI = dyn_cast<TruncInst>(Op0)) {
    Instruction *TrOp = dyn_cast<Instruction>(TI->getOperand(0));
    // If 'shift2' is an ashr, we would have to get the sign bit into a funny
    // place.  Don't try to do this transformation in this case.  Also, we
    // require that the input operand is a shift-by-constant so that we have
    // confidence that the shifts will get folded together.  We could do this
    // xform in more cases, but it is unlikely to be profitable.
    if (TrOp && I.isLogicalShift() && TrOp->isShift() &&
        isa<ConstantInt>(TrOp->getOperand(1))) {
      // Okay, we'll do this xform.  Make the shift of shift.
      Constant *ShAmt = ConstantExpr::getZExt(COp1, TrOp->getType());
      // (shift2 (shift1 & 0x00FF), c2)
      Value *NSh = Builder->CreateBinOp(I.getOpcode(), TrOp, ShAmt,I.getName());

      // For logical shifts, the truncation has the effect of making the high
      // part of the register be zeros.  Emulate this by inserting an AND to
      // clear the top bits as needed.  This 'and' will usually be zapped by
      // other xforms later if dead.
      unsigned SrcSize = TrOp->getType()->getScalarSizeInBits();
      unsigned DstSize = TI->getType()->getScalarSizeInBits();
      APInt MaskV(APInt::getLowBitsSet(SrcSize, DstSize));

      // The mask we constructed says what the trunc would do if occurring
      // between the shifts.  We want to know the effect *after* the second
      // shift.  We know that it is a logical shift by a constant, so adjust the
      // mask as appropriate.
      if (I.getOpcode() == Instruction::Shl)
        MaskV <<= COp1->getZExtValue();
      else {
        assert(I.getOpcode() == Instruction::LShr && "Unknown logical shift");
        MaskV = MaskV.lshr(COp1->getZExtValue());
      }

      // shift1 & 0x00FF
      Value *And = Builder->CreateAnd(NSh,
                                      ConstantInt::get(I.getContext(), MaskV),
                                      TI->getName());

      // Return the value truncated to the interesting size.
      return new TruncInst(And, I.getType());
    }
  }

  if (Op0->hasOneUse()) {
    if (BinaryOperator *Op0BO = dyn_cast<BinaryOperator>(Op0)) {
      // Turn ((X >> C) + Y) << C  ->  (X + (Y << C)) & (~0 << C)
      Value *V1, *V2;
      ConstantInt *CC;
      switch (Op0BO->getOpcode()) {
      default: break;
      case Instruction::Add:
      case Instruction::And:
      case Instruction::Or:
      case Instruction::Xor: {
        // These operators commute.
        // Turn (Y + (X >> C)) << C  ->  (X + (Y << C)) & (~0 << C)
        if (isLeftShift && Op0BO->getOperand(1)->hasOneUse() &&
            match(Op0BO->getOperand(1), m_Shr(m_Value(V1),
                  m_Specific(Op1)))) {
          Value *YS =         // (Y << C)
            Builder->CreateShl(Op0BO->getOperand(0), Op1, Op0BO->getName());
          // (X + (Y << C))
          Value *X = Builder->CreateBinOp(Op0BO->getOpcode(), YS, V1,
                                          Op0BO->getOperand(1)->getName());
          uint32_t Op1Val = COp1->getLimitedValue(TypeBits);

          APInt Bits = APInt::getHighBitsSet(TypeBits, TypeBits - Op1Val);
          Constant *Mask = ConstantInt::get(I.getContext(), Bits);
          if (VectorType *VT = dyn_cast<VectorType>(X->getType()))
            Mask = ConstantVector::getSplat(VT->getNumElements(), Mask);
          return BinaryOperator::CreateAnd(X, Mask);
        }

        // Turn (Y + ((X >> C) & CC)) << C  ->  ((X & (CC << C)) + (Y << C))
        Value *Op0BOOp1 = Op0BO->getOperand(1);
        if (isLeftShift && Op0BOOp1->hasOneUse() &&
            match(Op0BOOp1,
                  m_And(m_OneUse(m_Shr(m_Value(V1), m_Specific(Op1))),
                        m_ConstantInt(CC)))) {
          Value *YS =   // (Y << C)
            Builder->CreateShl(Op0BO->getOperand(0), Op1,
                                         Op0BO->getName());
          // X & (CC << C)
          Value *XM = Builder->CreateAnd(V1, ConstantExpr::getShl(CC, Op1),
                                         V1->getName()+".mask");
          return BinaryOperator::Create(Op0BO->getOpcode(), YS, XM);
        }
      }

      // FALL THROUGH.
      case Instruction::Sub: {
        // Turn ((X >> C) + Y) << C  ->  (X + (Y << C)) & (~0 << C)
        if (isLeftShift && Op0BO->getOperand(0)->hasOneUse() &&
            match(Op0BO->getOperand(0), m_Shr(m_Value(V1),
                  m_Specific(Op1)))) {
          Value *YS =  // (Y << C)
            Builder->CreateShl(Op0BO->getOperand(1), Op1, Op0BO->getName());
          // (X + (Y << C))
          Value *X = Builder->CreateBinOp(Op0BO->getOpcode(), V1, YS,
                                          Op0BO->getOperand(0)->getName());
          uint32_t Op1Val = COp1->getLimitedValue(TypeBits);

          APInt Bits = APInt::getHighBitsSet(TypeBits, TypeBits - Op1Val);
          Constant *Mask = ConstantInt::get(I.getContext(), Bits);
          if (VectorType *VT = dyn_cast<VectorType>(X->getType()))
            Mask = ConstantVector::getSplat(VT->getNumElements(), Mask);
          return BinaryOperator::CreateAnd(X, Mask);
        }

        // Turn (((X >> C)&CC) + Y) << C  ->  (X + (Y << C)) & (CC << C)
        if (isLeftShift && Op0BO->getOperand(0)->hasOneUse() &&
            match(Op0BO->getOperand(0),
                  m_And(m_OneUse(m_Shr(m_Value(V1), m_Value(V2))),
                        m_ConstantInt(CC))) && V2 == Op1) {
          Value *YS = // (Y << C)
            Builder->CreateShl(Op0BO->getOperand(1), Op1, Op0BO->getName());
          // X & (CC << C)
          Value *XM = Builder->CreateAnd(V1, ConstantExpr::getShl(CC, Op1),
                                         V1->getName()+".mask");

          return BinaryOperator::Create(Op0BO->getOpcode(), XM, YS);
        }

        break;
      }
      }


      // If the operand is an bitwise operator with a constant RHS, and the
      // shift is the only use, we can pull it out of the shift.
      if (ConstantInt *Op0C = dyn_cast<ConstantInt>(Op0BO->getOperand(1))) {
        bool isValid = true;     // Valid only for And, Or, Xor
        bool highBitSet = false; // Transform if high bit of constant set?

        switch (Op0BO->getOpcode()) {
        default: isValid = false; break;   // Do not perform transform!
        case Instruction::Add:
          isValid = isLeftShift;
          break;
        case Instruction::Or:
        case Instruction::Xor:
          highBitSet = false;
          break;
        case Instruction::And:
          highBitSet = true;
          break;
        }

        // If this is a signed shift right, and the high bit is modified
        // by the logical operation, do not perform the transformation.
        // The highBitSet boolean indicates the value of the high bit of
        // the constant which would cause it to be modified for this
        // operation.
        //
        if (isValid && I.getOpcode() == Instruction::AShr)
          isValid = Op0C->getValue()[TypeBits-1] == highBitSet;

        if (isValid) {
          Constant *NewRHS = ConstantExpr::get(I.getOpcode(), Op0C, Op1);

          Value *NewShift =
            Builder->CreateBinOp(I.getOpcode(), Op0BO->getOperand(0), Op1);
          NewShift->takeName(Op0BO);

          return BinaryOperator::Create(Op0BO->getOpcode(), NewShift,
                                        NewRHS);
        }
      }
    }
  }

  // Find out if this is a shift of a shift by a constant.
  BinaryOperator *ShiftOp = dyn_cast<BinaryOperator>(Op0);
  if (ShiftOp && !ShiftOp->isShift())
    ShiftOp = nullptr;

  if (ShiftOp && isa<ConstantInt>(ShiftOp->getOperand(1))) {

    // This is a constant shift of a constant shift. Be careful about hiding
    // shl instructions behind bit masks. They are used to represent multiplies
    // by a constant, and it is important that simple arithmetic expressions
    // are still recognizable by scalar evolution.
    //
    // The transforms applied to shl are very similar to the transforms applied
    // to mul by constant. We can be more aggressive about optimizing right
    // shifts.
    //
    // Combinations of right and left shifts will still be optimized in
    // DAGCombine where scalar evolution no longer applies.

    ConstantInt *ShiftAmt1C = cast<ConstantInt>(ShiftOp->getOperand(1));
    uint32_t ShiftAmt1 = ShiftAmt1C->getLimitedValue(TypeBits);
    uint32_t ShiftAmt2 = COp1->getLimitedValue(TypeBits);
    assert(ShiftAmt2 != 0 && "Should have been simplified earlier");
    if (ShiftAmt1 == 0) return nullptr;  // Will be simplified in the future.
    Value *X = ShiftOp->getOperand(0);

    IntegerType *Ty = cast<IntegerType>(I.getType());

    // Check for (X << c1) << c2  and  (X >> c1) >> c2
    if (I.getOpcode() == ShiftOp->getOpcode()) {
      uint32_t AmtSum = ShiftAmt1+ShiftAmt2;   // Fold into one big shift.
      // If this is oversized composite shift, then unsigned shifts get 0, ashr
      // saturates.
      if (AmtSum >= TypeBits) {
        if (I.getOpcode() != Instruction::AShr)
          return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
        AmtSum = TypeBits-1;  // Saturate to 31 for i32 ashr.
      }

      return BinaryOperator::Create(I.getOpcode(), X,
                                    ConstantInt::get(Ty, AmtSum));
    }

    if (ShiftAmt1 == ShiftAmt2) {
      // If we have ((X << C) >>u C), turn this into X & (-1 >>u C).
      if (I.getOpcode() == Instruction::LShr &&
          ShiftOp->getOpcode() == Instruction::Shl) {
        APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt1));
        return BinaryOperator::CreateAnd(X,
                                        ConstantInt::get(I.getContext(), Mask));
      }
    } else if (ShiftAmt1 < ShiftAmt2) {
      uint32_t ShiftDiff = ShiftAmt2-ShiftAmt1;

      // (X >>?,exact C1) << C2 --> X << (C2-C1)
      // The inexact version is deferred to DAGCombine so we don't hide shl
      // behind a bit mask.
      if (I.getOpcode() == Instruction::Shl &&
          ShiftOp->getOpcode() != Instruction::Shl &&
          ShiftOp->isExact()) {
        assert(ShiftOp->getOpcode() == Instruction::LShr ||
               ShiftOp->getOpcode() == Instruction::AShr);
        ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
        BinaryOperator *NewShl = BinaryOperator::Create(Instruction::Shl,
                                                        X, ShiftDiffCst);
        NewShl->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
        NewShl->setHasNoSignedWrap(I.hasNoSignedWrap());
        return NewShl;
      }

      // (X << C1) >>u C2  --> X >>u (C2-C1) & (-1 >> C2)
      if (I.getOpcode() == Instruction::LShr &&
          ShiftOp->getOpcode() == Instruction::Shl) {
        ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
        // (X <<nuw C1) >>u C2 --> X >>u (C2-C1)
        if (ShiftOp->hasNoUnsignedWrap()) {
          BinaryOperator *NewLShr = BinaryOperator::Create(Instruction::LShr,
                                                           X, ShiftDiffCst);
          NewLShr->setIsExact(I.isExact());
          return NewLShr;
        }
        Value *Shift = Builder->CreateLShr(X, ShiftDiffCst);

        APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt2));
        return BinaryOperator::CreateAnd(Shift,
                                         ConstantInt::get(I.getContext(),Mask));
      }

      // We can't handle (X << C1) >>s C2, it shifts arbitrary bits in. However,
      // we can handle (X <<nsw C1) >>s C2 since it only shifts in sign bits.
      if (I.getOpcode() == Instruction::AShr &&
          ShiftOp->getOpcode() == Instruction::Shl) {
        if (ShiftOp->hasNoSignedWrap()) {
          // (X <<nsw C1) >>s C2 --> X >>s (C2-C1)
          ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
          BinaryOperator *NewAShr = BinaryOperator::Create(Instruction::AShr,
                                                           X, ShiftDiffCst);
          NewAShr->setIsExact(I.isExact());
          return NewAShr;
        }
      }
    } else {
      assert(ShiftAmt2 < ShiftAmt1);
      uint32_t ShiftDiff = ShiftAmt1-ShiftAmt2;

      // (X >>?exact C1) << C2 --> X >>?exact (C1-C2)
      // The inexact version is deferred to DAGCombine so we don't hide shl
      // behind a bit mask.
      if (I.getOpcode() == Instruction::Shl &&
          ShiftOp->getOpcode() != Instruction::Shl &&
          ShiftOp->isExact()) {
        ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
        BinaryOperator *NewShr = BinaryOperator::Create(ShiftOp->getOpcode(),
                                                        X, ShiftDiffCst);
        NewShr->setIsExact(true);
        return NewShr;
      }

      // (X << C1) >>u C2  --> X << (C1-C2) & (-1 >> C2)
      if (I.getOpcode() == Instruction::LShr &&
          ShiftOp->getOpcode() == Instruction::Shl) {
        ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
        if (ShiftOp->hasNoUnsignedWrap()) {
          // (X <<nuw C1) >>u C2 --> X <<nuw (C1-C2)
          BinaryOperator *NewShl = BinaryOperator::Create(Instruction::Shl,
                                                          X, ShiftDiffCst);
          NewShl->setHasNoUnsignedWrap(true);
          return NewShl;
        }
        Value *Shift = Builder->CreateShl(X, ShiftDiffCst);

        APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt2));
        return BinaryOperator::CreateAnd(Shift,
                                         ConstantInt::get(I.getContext(),Mask));
      }

      // We can't handle (X << C1) >>s C2, it shifts arbitrary bits in. However,
      // we can handle (X <<nsw C1) >>s C2 since it only shifts in sign bits.
      if (I.getOpcode() == Instruction::AShr &&
          ShiftOp->getOpcode() == Instruction::Shl) {
        if (ShiftOp->hasNoSignedWrap()) {
          // (X <<nsw C1) >>s C2 --> X <<nsw (C1-C2)
          ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
          BinaryOperator *NewShl = BinaryOperator::Create(Instruction::Shl,
                                                          X, ShiftDiffCst);
          NewShl->setHasNoSignedWrap(true);
          return NewShl;
        }
      }
    }
  }
  return nullptr;
}

Instruction * InstCombiner::FoldSPFofSPF	(	Instruction *	Inner,
		SelectPatternFlavor	SPF1,
		Value *	A,
		Value *	B,
		Instruction &	Outer,
		SelectPatternFlavor	SPF2,
		Value *	C
	)

FoldSPFofSPF - We have an SPF (e.g. a min or max) of an SPF of the form: SPF2(SPF1(A, B), C)

Definition at line 683 of file InstCombineSelect.cpp.

                                                                            {
  if (C == A || C == B) {
    // MAX(MAX(A, B), B) -> MAX(A, B)
    // MIN(MIN(a, b), a) -> MIN(a, b)
    if (SPF1 == SPF2)
      return ReplaceInstUsesWith(Outer, Inner);

    // MAX(MIN(a, b), a) -> a
    // MIN(MAX(a, b), a) -> a
    if ((SPF1 == SPF_SMIN && SPF2 == SPF_SMAX) ||
        (SPF1 == SPF_SMAX && SPF2 == SPF_SMIN) ||
        (SPF1 == SPF_UMIN && SPF2 == SPF_UMAX) ||
        (SPF1 == SPF_UMAX && SPF2 == SPF_UMIN))
      return ReplaceInstUsesWith(Outer, C);
  }

  if (SPF1 == SPF2) {
    if (ConstantInt *CB = dyn_cast<ConstantInt>(B)) {
      if (ConstantInt *CC = dyn_cast<ConstantInt>(C)) {
        APInt ACB = CB->getValue();
        APInt ACC = CC->getValue();

        // MIN(MIN(A, 23), 97) -> MIN(A, 23)
        // MAX(MAX(A, 97), 23) -> MAX(A, 97)
        if ((SPF1 == SPF_UMIN && ACB.ule(ACC)) ||
            (SPF1 == SPF_SMIN && ACB.sle(ACC)) ||
            (SPF1 == SPF_UMAX && ACB.uge(ACC)) ||
            (SPF1 == SPF_SMAX && ACB.sge(ACC)))
          return ReplaceInstUsesWith(Outer, Inner);

        // MIN(MIN(A, 97), 23) -> MIN(A, 23)
        // MAX(MAX(A, 23), 97) -> MAX(A, 97)
        if ((SPF1 == SPF_UMIN && ACB.ugt(ACC)) ||
            (SPF1 == SPF_SMIN && ACB.sgt(ACC)) ||
            (SPF1 == SPF_UMAX && ACB.ult(ACC)) ||
            (SPF1 == SPF_SMAX && ACB.slt(ACC))) {
          Outer.replaceUsesOfWith(Inner, A);
          return &Outer;
        }
      }
    }
  }

  // ABS(ABS(X)) -> ABS(X)
  // NABS(NABS(X)) -> NABS(X)
  if (SPF1 == SPF2 && (SPF1 == SPF_ABS || SPF1 == SPF_NABS)) {
    return ReplaceInstUsesWith(Outer, Inner);
  }

  // ABS(NABS(X)) -> ABS(X)
  // NABS(ABS(X)) -> NABS(X)
  if ((SPF1 == SPF_ABS && SPF2 == SPF_NABS) ||
      (SPF1 == SPF_NABS && SPF2 == SPF_ABS)) {
    SelectInst *SI = cast<SelectInst>(Inner);
    Value *NewSI = Builder->CreateSelect(
        SI->getCondition(), SI->getFalseValue(), SI->getTrueValue());
    return ReplaceInstUsesWith(Outer, NewSI);
  }
  return nullptr;
}

void * Pass::getAdjustedAnalysisPointer ( AnalysisID  ID )

virtualinherited

getAdjustedAnalysisPointer - This method is used when a pass implements an analysis interface through multiple inheritance. If needed, it should override this to adjust the this pointer as needed for the specified pass info.

Reimplemented in llvm::LibCallAliasAnalysis, and llvm::objcarc::ObjCARCAliasAnalysis.

Definition at line 90 of file Pass.cpp.

                                                     {
  return this;
}

template<typename AnalysisType >

AnalysisType & llvm::Pass::getAnalysis ( ) const

inherited

getAnalysis<AnalysisType>() - This function is used by subclasses to get to the analysis information that they claim to use by overriding the getAnalysisUsage function.

Definition at line 199 of file PassAnalysisSupport.h.

                                      {
  assert(Resolver && "Pass has not been inserted into a PassManager object!");
  return getAnalysisID<AnalysisType>(&AnalysisType::ID);
}

template<typename AnalysisType >

AnalysisType & llvm::Pass::getAnalysis ( Function &  F )

inherited

getAnalysis<AnalysisType>() - This function is used by subclasses to get to the analysis information that they claim to use by overriding the getAnalysisUsage function.

Definition at line 228 of file PassAnalysisSupport.h.

                                           {
  assert(Resolver &&"Pass has not been inserted into a PassManager object!");

  return getAnalysisID<AnalysisType>(&AnalysisType::ID, F);
}

template<typename AnalysisType >

AnalysisType & llvm::Pass::getAnalysisID ( AnalysisID  PI ) const

inherited

Definition at line 205 of file PassAnalysisSupport.h.

                                                     {
  assert(PI && "getAnalysis for unregistered pass!");
  assert(Resolver&&"Pass has not been inserted into a PassManager object!");
  // PI *must* appear in AnalysisImpls.  Because the number of passes used
  // should be a small number, we just do a linear search over a (dense)
  // vector.
  Pass *ResultPass = Resolver->findImplPass(PI);
  assert (ResultPass && 
          "getAnalysis*() called on an analysis that was not "
          "'required' by pass!");

  // Because the AnalysisType may not be a subclass of pass (for
  // AnalysisGroups), we use getAdjustedAnalysisPointer here to potentially
  // adjust the return pointer (because the class may multiply inherit, once
  // from pass, once from AnalysisType).
  return *(AnalysisType*)ResultPass->getAdjustedAnalysisPointer(PI);
}

template<typename AnalysisType >

AnalysisType & llvm::Pass::getAnalysisID ( AnalysisID  PI, Function &  F  )

inherited

Definition at line 235 of file PassAnalysisSupport.h.

                                                            {
  assert(PI && "getAnalysis for unregistered pass!");
  assert(Resolver && "Pass has not been inserted into a PassManager object!");
  // PI *must* appear in AnalysisImpls.  Because the number of passes used
  // should be a small number, we just do a linear search over a (dense)
  // vector.
  Pass *ResultPass = Resolver->findImplPass(this, PI, F);
  assert(ResultPass && "Unable to find requested analysis info");
  
  // Because the AnalysisType may not be a subclass of pass (for
  // AnalysisGroups), we use getAdjustedAnalysisPointer here to potentially
  // adjust the return pointer (because the class may multiply inherit, once
  // from pass, once from AnalysisType).
  return *(AnalysisType*)ResultPass->getAdjustedAnalysisPointer(PI);
}

template<typename AnalysisType >

AnalysisType * llvm::Pass::getAnalysisIfAvailable ( ) const

inherited

getAnalysisIfAvailable<AnalysisType>() - Subclasses use this function to get analysis information that might be around, for example to update it. This is different than getAnalysis in that it can fail (if the analysis results haven't been computed), so should only be used if you can handle the case when the analysis is not available. This method is often used by transformation APIs to update analysis results for a pass automatically as the transform is performed.

Definition at line 179 of file PassAnalysisSupport.h.

                                                 {
  assert(Resolver && "Pass not resident in a PassManager object!");

  const void *PI = &AnalysisType::ID;

  Pass *ResultPass = Resolver->getAnalysisIfAvailable(PI, true);
  if (!ResultPass) return nullptr;

  // Because the AnalysisType may not be a subclass of pass (for
  // AnalysisGroups), we use getAdjustedAnalysisPointer here to potentially
  // adjust the return pointer (because the class may multiply inherit, once
  // from pass, once from AnalysisType).
  return (AnalysisType*)ResultPass->getAdjustedAnalysisPointer(PI);
}

Combine redundant false void InstCombiner::getAnalysisUsage ( AnalysisUsage &  AU ) const

overridevirtual

Reimplemented from llvm::Pass.

Definition at line 92 of file InstructionCombining.cpp.

                                                           {
  AU.setPreservesCFG();
  AU.addRequired<TargetLibraryInfo>();
}

ImmutablePass * Pass::getAsImmutablePass ( )

virtualinherited

Reimplemented in llvm::ImmutablePass.

Definition at line 94 of file Pass.cpp.

                                        {
  return nullptr;
}

PMDataManager * Pass::getAsPMDataManager ( )

virtualinherited

Reimplemented in llvm::FPPassManager, llvm::legacy::PassManagerImpl, llvm::legacy::FunctionPassManagerImpl, llvm::LPPassManager, and llvm::RGPassManager.

Definition at line 98 of file Pass.cpp.

                                        {
  return nullptr;
}

const DataLayout* llvm::InstCombiner::getDataLayout ( ) const

inline

Definition at line 117 of file InstCombine.h.

{ return DL; }

AnalysisID llvm::Pass::getPassID ( ) const

inlineinherited

getPassID - Return the PassID number that corresponds to this pass.

Definition at line 104 of file Pass.h.

                               {
    return PassID;
  }

PassKind llvm::Pass::getPassKind ( ) const

inlineinherited

Definition at line 95 of file Pass.h.

{ return Kind; }

const char * Pass::getPassName ( ) const

virtualinherited

getPassName - Return a nice clean name for a pass. This usually implemented in terms of the name that is registered by one of the Registration templates, but can be overloaded directly.

Reimplemented in llvm::FPPassManager, llvm::NVPTXAsmPrinter, llvm::LPPassManager, llvm::RGPassManager, llvm::MipsAsmPrinter, llvm::AMDGPUAsmPrinter, llvm::JumpInstrTables, llvm::ARMAsmPrinter, legup::LegupPass, llvm::DebugIR, llvm::MipsModuleDAGToDAGISel, llvm::JumpInstrTableInfo, llvm::NVPTXLowerAggrCopies, llvm::NVPTXAllocaHoisting, llvm::MachineFunctionAnalysis, llvm::MipsDAGToDAGISel, llvm::Mips16HardFloat, llvm::MipsOs16, llvm::SystemZAsmPrinter, llvm::X86AsmPrinter, and llvm::HexagonAsmPrinter.

Definition at line 61 of file Pass.cpp.

                                    {
  AnalysisID AID =  getPassID();
  const PassInfo *PI = PassRegistry::getPassRegistry()->getPassInfo(AID);
  if (PI)
    return PI->getPassName();
  return "Unnamed pass: implement Pass::getPassName()";
}

PassManagerType FunctionPass::getPotentialPassManagerType ( ) const

overridevirtualinherited

Return what kind of Pass Manager can manage this pass.

Reimplemented from llvm::Pass.

Definition at line 139 of file Pass.cpp.

                                                                {
  return PMT_FunctionPassManager;
}

AnalysisResolver* llvm::Pass::getResolver ( ) const

inlineinherited

Definition at line 145 of file Pass.h.

{ return Resolver; }

TargetLibraryInfo* llvm::InstCombiner::getTargetLibraryInfo ( ) const

inline

Definition at line 119 of file InstCombine.h.

{ return TLI; }

Instruction* llvm::InstCombiner::InsertNewInstBefore ( Instruction *  New, Instruction &  Old  )

inline

Definition at line 259 of file InstCombine.h.

                                                                       {
    assert(New && !New->getParent() &&
           "New instruction already inserted into a basic block!");
    BasicBlock *BB = Old.getParent();
    BB->getInstList().insert(&Old, New); // Insert inst
    Worklist.Add(New);
    return New;
  }

Instruction* llvm::InstCombiner::InsertNewInstWith ( Instruction *  New, Instruction &  Old  )

inline

Definition at line 271 of file InstCombine.h.

                                                                     {
    New->setDebugLoc(Old.getDebugLoc());
    return InsertNewInstBefore(New, Old);
  }

Value * InstCombiner::InsertRangeTest ( Value *  V, Constant *  Lo, Constant *  Hi, bool  isSigned, bool  Inside  )

private

Emit a computation of: (V >= Lo && V < Hi) if Inside is true, otherwise (V < Lo || V >= Hi). In practice, we emit the more efficient (V-Lo) <u Hi-Lo. This method expects that Lo <= Hi. isSigned indicates whether to treat the V, Lo and HI as signed or not. IB is the location to insert new instructions.

Definition at line 263 of file InstCombineAndOrXor.cpp.

                                                                 {
  assert(cast<ConstantInt>(ConstantExpr::getICmp((isSigned ?
            ICmpInst::ICMP_SLE:ICmpInst::ICMP_ULE), Lo, Hi))->getZExtValue() &&
         "Lo is not <= Hi in range emission code!");

  if (Inside) {
    if (Lo == Hi)  // Trivially false.
      return Builder->getFalse();

    // V >= Min && V < Hi --> V < Hi
    if (cast<ConstantInt>(Lo)->isMinValue(isSigned)) {
      ICmpInst::Predicate pred = (isSigned ?
        ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT);
      return Builder->CreateICmp(pred, V, Hi);
    }

    // Emit V-Lo <u Hi-Lo
    Constant *NegLo = ConstantExpr::getNeg(Lo);
    Value *Add = Builder->CreateAdd(V, NegLo, V->getName()+".off");
    Constant *UpperBound = ConstantExpr::getAdd(NegLo, Hi);
    return Builder->CreateICmpULT(Add, UpperBound);
  }

  if (Lo == Hi)  // Trivially true.
    return Builder->getTrue();

  // V < Min || V >= Hi -> V > Hi-1
  Hi = SubOne(cast<ConstantInt>(Hi));
  if (cast<ConstantInt>(Lo)->isMinValue(isSigned)) {
    ICmpInst::Predicate pred = (isSigned ?
        ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT);
    return Builder->CreateICmp(pred, V, Hi);
  }

  // Emit V-Lo >u Hi-1-Lo
  // Note that Hi has already had one subtracted from it, above.
  ConstantInt *NegLo = cast<ConstantInt>(ConstantExpr::getNeg(Lo));
  Value *Add = Builder->CreateAdd(V, NegLo, V->getName()+".off");
  Constant *LowerBound = ConstantExpr::getAdd(NegLo, Hi);
  return Builder->CreateICmpUGT(Add, LowerBound);
}

const PassInfo * Pass::lookupPassInfo ( const void *  TI )

staticinherited

Definition at line 187 of file Pass.cpp.

                                                   {
  return PassRegistry::getPassRegistry()->getPassInfo(TI);
}

const PassInfo * Pass::lookupPassInfo ( StringRef  Arg )

staticinherited

Definition at line 191 of file Pass.cpp.

                                                  {
  return PassRegistry::getPassRegistry()->getPassInfo(Arg);
}

bool llvm::InstCombiner::MaskedValueIsZero ( Value *  V, const APInt &  Mask, unsigned  Depth = 0  ) const

inline

Definition at line 323 of file InstCombine.h.

                                                   {
    return llvm::MaskedValueIsZero(V, Mask, DL, Depth);
  }

Instruction * InstCombiner::MatchBSwap ( BinaryOperator &  I )

private

MatchBSwap - Given an OR instruction, check to see if this is a bswap idiom. If so, insert the new bswap intrinsic and return it.

ByteValues - For each byte of the result, we keep track of which value defines each byte.

Definition at line 1501 of file InstCombineAndOrXor.cpp.

                                                       {
  IntegerType *ITy = dyn_cast<IntegerType>(I.getType());
  if (!ITy || ITy->getBitWidth() % 16 ||
      // ByteMask only allows up to 32-byte values.
      ITy->getBitWidth() > 32*8)
    return nullptr;   // Can only bswap pairs of bytes.  Can't do vectors.

  /// ByteValues - For each byte of the result, we keep track of which value
  /// defines each byte.
  SmallVector<Value*, 8> ByteValues;
  ByteValues.resize(ITy->getBitWidth()/8);

  // Try to find all the pieces corresponding to the bswap.
  uint32_t ByteMask = ~0U >> (32-ByteValues.size());
  if (CollectBSwapParts(&I, 0, ByteMask, ByteValues))
    return nullptr;

  // Check to see if all of the bytes come from the same value.
  Value *V = ByteValues[0];
  if (!V) return nullptr;  // Didn't find a byte?  Must be zero.

  // Check to make sure that all of the bytes come from the same value.
  for (unsigned i = 1, e = ByteValues.size(); i != e; ++i)
    if (ByteValues[i] != V)
      return nullptr;
  Module *M = I.getParent()->getParent()->getParent();
  Function *F = Intrinsic::getDeclaration(M, Intrinsic::bswap, ITy);
  return CallInst::Create(F, V);
}

bool Pass::mustPreserveAnalysisID ( char &  AID ) const

inherited

mustPreserveAnalysisID - This method serves the same function as getAnalysisIfAvailable, but works if you just have an AnalysisID. This obviously cannot give you a properly typed instance of the class if you don't have the class name available (use getAnalysisIfAvailable if you do), but it can tell you if you need to preserve the pass at least.

Definition at line 48 of file Pass.cpp.

                                                 {
  return Resolver->getAnalysisIfAvailable(&AID, true) != nullptr;
}

Instruction * InstCombiner::OptAndOp ( Instruction *  Op, ConstantInt *  OpRHS, ConstantInt *  AndRHS, BinaryOperator &  TheAnd  )

private

Definition at line 123 of file InstCombineAndOrXor.cpp.

                                                            {
  Value *X = Op->getOperand(0);
  Constant *Together = nullptr;
  if (!Op->isShift())
    Together = ConstantExpr::getAnd(AndRHS, OpRHS);

  switch (Op->getOpcode()) {
  case Instruction::Xor:
    if (Op->hasOneUse()) {
      // (X ^ C1) & C2 --> (X & C2) ^ (C1&C2)
      Value *And = Builder->CreateAnd(X, AndRHS);
      And->takeName(Op);
      return BinaryOperator::CreateXor(And, Together);
    }
    break;
  case Instruction::Or:
    if (Op->hasOneUse()){
      if (Together != OpRHS) {
        // (X | C1) & C2 --> (X | (C1&C2)) & C2
        Value *Or = Builder->CreateOr(X, Together);
        Or->takeName(Op);
        return BinaryOperator::CreateAnd(Or, AndRHS);
      }

      ConstantInt *TogetherCI = dyn_cast<ConstantInt>(Together);
      if (TogetherCI && !TogetherCI->isZero()){
        // (X | C1) & C2 --> (X & (C2^(C1&C2))) | C1
        // NOTE: This reduces the number of bits set in the & mask, which
        // can expose opportunities for store narrowing.
        Together = ConstantExpr::getXor(AndRHS, Together);
        Value *And = Builder->CreateAnd(X, Together);
        And->takeName(Op);
        return BinaryOperator::CreateOr(And, OpRHS);
      }
    }

    break;
  case Instruction::Add:
    if (Op->hasOneUse()) {
      // Adding a one to a single bit bit-field should be turned into an XOR
      // of the bit.  First thing to check is to see if this AND is with a
      // single bit constant.
      const APInt &AndRHSV = AndRHS->getValue();

      // If there is only one bit set.
      if (AndRHSV.isPowerOf2()) {
        // Ok, at this point, we know that we are masking the result of the
        // ADD down to exactly one bit.  If the constant we are adding has
        // no bits set below this bit, then we can eliminate the ADD.
        const APInt& AddRHS = OpRHS->getValue();

        // Check to see if any bits below the one bit set in AndRHSV are set.
        if ((AddRHS & (AndRHSV-1)) == 0) {
          // If not, the only thing that can effect the output of the AND is
          // the bit specified by AndRHSV.  If that bit is set, the effect of
          // the XOR is to toggle the bit.  If it is clear, then the ADD has
          // no effect.
          if ((AddRHS & AndRHSV) == 0) { // Bit is not set, noop
            TheAnd.setOperand(0, X);
            return &TheAnd;
          } else {
            // Pull the XOR out of the AND.
            Value *NewAnd = Builder->CreateAnd(X, AndRHS);
            NewAnd->takeName(Op);
            return BinaryOperator::CreateXor(NewAnd, AndRHS);
          }
        }
      }
    }
    break;

  case Instruction::Shl: {
    // We know that the AND will not produce any of the bits shifted in, so if
    // the anded constant includes them, clear them now!
    //
    uint32_t BitWidth = AndRHS->getType()->getBitWidth();
    uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
    APInt ShlMask(APInt::getHighBitsSet(BitWidth, BitWidth-OpRHSVal));
    ConstantInt *CI = Builder->getInt(AndRHS->getValue() & ShlMask);

    if (CI->getValue() == ShlMask)
      // Masking out bits that the shift already masks.
      return ReplaceInstUsesWith(TheAnd, Op);   // No need for the and.

    if (CI != AndRHS) {                  // Reducing bits set in and.
      TheAnd.setOperand(1, CI);
      return &TheAnd;
    }
    break;
  }
  case Instruction::LShr: {
    // We know that the AND will not produce any of the bits shifted in, so if
    // the anded constant includes them, clear them now!  This only applies to
    // unsigned shifts, because a signed shr may bring in set bits!
    //
    uint32_t BitWidth = AndRHS->getType()->getBitWidth();
    uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
    APInt ShrMask(APInt::getLowBitsSet(BitWidth, BitWidth - OpRHSVal));
    ConstantInt *CI = Builder->getInt(AndRHS->getValue() & ShrMask);

    if (CI->getValue() == ShrMask)
      // Masking out bits that the shift already masks.
      return ReplaceInstUsesWith(TheAnd, Op);

    if (CI != AndRHS) {
      TheAnd.setOperand(1, CI);  // Reduce bits set in and cst.
      return &TheAnd;
    }
    break;
  }
  case Instruction::AShr:
    // Signed shr.
    // See if this is shifting in some sign extension, then masking it out
    // with an and.
    if (Op->hasOneUse()) {
      uint32_t BitWidth = AndRHS->getType()->getBitWidth();
      uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
      APInt ShrMask(APInt::getLowBitsSet(BitWidth, BitWidth - OpRHSVal));
      Constant *C = Builder->getInt(AndRHS->getValue() & ShrMask);
      if (C == AndRHS) {          // Masking out bits shifted in.
        // (Val ashr C1) & C2 -> (Val lshr C1) & C2
        // Make the argument unsigned.
        Value *ShVal = Op->getOperand(0);
        ShVal = Builder->CreateLShr(ShVal, OpRHS, Op->getName());
        return BinaryOperator::CreateAnd(ShVal, AndRHS, TheAnd.getName());
      }
    }
    break;
  }
  return nullptr;
}

Value * InstCombiner::OptimizePointerDifference	(	Value *	LHS,
		Value *	RHS,
		Type *	Ty
	)

Optimize pointer differences into the same array into a size. Consider: &A[10] - &A[0]: we should compile this to "10". LHS/RHS are the pointer operands to the ptrtoint instructions for the LHS/RHS of the subtract.

Definition at line 1384 of file InstCombineAddSub.cpp.

                                                         {
  assert(DL && "Must have target data info for this");

  // If LHS is a gep based on RHS or RHS is a gep based on LHS, we can optimize
  // this.
  bool Swapped = false;
  GEPOperator *GEP1 = nullptr, *GEP2 = nullptr;

  // For now we require one side to be the base pointer "A" or a constant
  // GEP derived from it.
  if (GEPOperator *LHSGEP = dyn_cast<GEPOperator>(LHS)) {
    // (gep X, ...) - X
    if (LHSGEP->getOperand(0) == RHS) {
      GEP1 = LHSGEP;
      Swapped = false;
    } else if (GEPOperator *RHSGEP = dyn_cast<GEPOperator>(RHS)) {
      // (gep X, ...) - (gep X, ...)
      if (LHSGEP->getOperand(0)->stripPointerCasts() ==
            RHSGEP->getOperand(0)->stripPointerCasts()) {
        GEP2 = RHSGEP;
        GEP1 = LHSGEP;
        Swapped = false;
      }
    }
  }

  if (GEPOperator *RHSGEP = dyn_cast<GEPOperator>(RHS)) {
    // X - (gep X, ...)
    if (RHSGEP->getOperand(0) == LHS) {
      GEP1 = RHSGEP;
      Swapped = true;
    } else if (GEPOperator *LHSGEP = dyn_cast<GEPOperator>(LHS)) {
      // (gep X, ...) - (gep X, ...)
      if (RHSGEP->getOperand(0)->stripPointerCasts() ==
            LHSGEP->getOperand(0)->stripPointerCasts()) {
        GEP2 = LHSGEP;
        GEP1 = RHSGEP;
        Swapped = true;
      }
    }
  }

  // Avoid duplicating the arithmetic if GEP2 has non-constant indices and
  // multiple users.
  if (!GEP1 ||
      (GEP2 && !GEP2->hasAllConstantIndices() && !GEP2->hasOneUse()))
    return nullptr;

  // Emit the offset of the GEP and an intptr_t.
  Value *Result = EmitGEPOffset(GEP1);

  // If we had a constant expression GEP on the other side offsetting the
  // pointer, subtract it from the offset we have.
  if (GEP2) {
    Value *Offset = EmitGEPOffset(GEP2);
    Result = Builder->CreateSub(Result, Offset);
  }

  // If we have p - gep(p, ...)  then we have to negate the result.
  if (Swapped)
    Result = Builder->CreateNeg(Result, "diff.neg");

  return Builder->CreateIntCast(Result, Ty, true);
}

void Pass::preparePassManager ( PMStack &  )

virtualinherited

Check if available pass managers are suitable for this pass or not.

Reimplemented in llvm::RegionPass, and llvm::LoopPass.

Definition at line 69 of file Pass.cpp.

                                       {
  // By default, don't do anything.
}

void Pass::print ( raw_ostream &  O, const Module *  M  ) const

virtualinherited

print - Print out the internal state of the pass. This is called by Analyze to print out the contents of an analysis. Otherwise it is not necessary to implement this method. Beware that the module pointer MAY be null. This automatically forwards to a virtual function that does not provide the Module* in case the analysis doesn't need it it can just be ignored.

Reimplemented in llvm::DependenceAnalysis, llvm::ScalarEvolution, llvm::RegionInfoPass, llvm::LoopInfo, llvm::CallGraphWrapperPass, llvm::LiveIntervals, llvm::DominanceFrontier, llvm::DominatorTreeWrapperPass, llvm::VirtRegMap, llvm::IVUsers, llvm::MachineDominatorTree, llvm::MachineRegionInfoPass, llvm::LiveStacks, llvm::PostDominatorTree, llvm::MachinePostDominatorTree, llvm::IntervalPartition, llvm::BranchProbabilityInfo, llvm::BlockFrequencyInfo, and llvm::FindUsedTypes.

Definition at line 111 of file Pass.cpp.

                                                   {
  O << "Pass::print not implemented for pass: '" << getPassName() << "'!\n";
}

Instruction * InstCombiner::PromoteCastOfAllocation ( BitCastInst &  CI, AllocaInst &  AI  )

private

PromoteCastOfAllocation - If we find a cast of an allocation instruction, try to eliminate the cast by moving the type information into the alloc.

Definition at line 81 of file InstCombineCasts.cpp.

                                                                   {
  // This requires DataLayout to get the alloca alignment and size information.
  if (!DL) return nullptr;

  PointerType *PTy = cast<PointerType>(CI.getType());

  BuilderTy AllocaBuilder(*Builder);
  AllocaBuilder.SetInsertPoint(AI.getParent(), &AI);

  // Get the type really allocated and the type casted to.
  Type *AllocElTy = AI.getAllocatedType();
  Type *CastElTy = PTy->getElementType();
  if (!AllocElTy->isSized() || !CastElTy->isSized()) return nullptr;

  unsigned AllocElTyAlign = DL->getABITypeAlignment(AllocElTy);
  unsigned CastElTyAlign = DL->getABITypeAlignment(CastElTy);
  if (CastElTyAlign < AllocElTyAlign) return nullptr;

  // If the allocation has multiple uses, only promote it if we are strictly
  // increasing the alignment of the resultant allocation.  If we keep it the
  // same, we open the door to infinite loops of various kinds.
  if (!AI.hasOneUse() && CastElTyAlign == AllocElTyAlign) return nullptr;

  uint64_t AllocElTySize = DL->getTypeAllocSize(AllocElTy);
  uint64_t CastElTySize = DL->getTypeAllocSize(CastElTy);
  if (CastElTySize == 0 || AllocElTySize == 0) return nullptr;

  // If the allocation has multiple uses, only promote it if we're not
  // shrinking the amount of memory being allocated.
  uint64_t AllocElTyStoreSize = DL->getTypeStoreSize(AllocElTy);
  uint64_t CastElTyStoreSize = DL->getTypeStoreSize(CastElTy);
  if (!AI.hasOneUse() && CastElTyStoreSize < AllocElTyStoreSize) return nullptr;

  // See if we can satisfy the modulus by pulling a scale out of the array
  // size argument.
  unsigned ArraySizeScale;
  uint64_t ArrayOffset;
  Value *NumElements = // See if the array size is a decomposable linear expr.
    DecomposeSimpleLinearExpr(AI.getOperand(0), ArraySizeScale, ArrayOffset);

  // If we can now satisfy the modulus, by using a non-1 scale, we really can
  // do the xform.
  if ((AllocElTySize*ArraySizeScale) % CastElTySize != 0 ||
      (AllocElTySize*ArrayOffset   ) % CastElTySize != 0) return nullptr;

  unsigned Scale = (AllocElTySize*ArraySizeScale)/CastElTySize;
  Value *Amt = nullptr;
  if (Scale == 1) {
    Amt = NumElements;
  } else {
    Amt = ConstantInt::get(AI.getArraySize()->getType(), Scale);
    // Insert before the alloca, not before the cast.
    Amt = AllocaBuilder.CreateMul(Amt, NumElements);
  }

  if (uint64_t Offset = (AllocElTySize*ArrayOffset)/CastElTySize) {
    Value *Off = ConstantInt::get(AI.getArraySize()->getType(),
                                  Offset, true);
    Amt = AllocaBuilder.CreateAdd(Amt, Off);
  }

  AllocaInst *New = AllocaBuilder.CreateAlloca(CastElTy, Amt);
  New->setAlignment(AI.getAlignment());
  New->takeName(&AI);
  New->setUsedWithInAlloca(AI.isUsedWithInAlloca());

  // If the allocation has multiple real uses, insert a cast and change all
  // things that used it to use the new cast.  This will also hack on CI, but it
  // will die soon.
  if (!AI.hasOneUse()) {
    // New is the allocation instruction, pointer typed. AI is the original
    // allocation instruction, also pointer typed. Thus, cast to use is BitCast.
    Value *NewCast = AllocaBuilder.CreateBitCast(New, AI.getType(), "tmpcast");
    ReplaceInstUsesWith(AI, NewCast);
  }
  return ReplaceInstUsesWith(CI, New);
}

void Pass::releaseMemory ( )

virtualinherited

releaseMemory() - This member can be implemented by a pass if it wants to be able to release its memory when it is no longer needed. The default behavior of passes is to hold onto memory for the entire duration of their lifetime (which is the entire compile time). For pipelined passes, this is not a big deal because that memory gets recycled every time the pass is invoked on another program unit. For IP passes, it is more important to free memory when it is unused.

Optionally implement this function to release pass memory when it is no longer used.

Reimplemented in llvm::DependenceAnalysis, llvm::ScalarEvolution, llvm::RegionInfoPass, llvm::LoopInfo, llvm::CallGraphWrapperPass, llvm::SlotIndexes, llvm::MemoryDependenceAnalysis, llvm::LiveIntervals, llvm::LiveVariables, llvm::DominanceFrontier, llvm::DominatorTreeWrapperPass, llvm::MachineDominatorTree, llvm::SpillPlacement, llvm::IVUsers, llvm::MachineLoopInfo, llvm::MachineRegionInfoPass, llvm::MachineDominanceFrontier, llvm::IntervalPartition, llvm::LiveStacks, llvm::PostDominatorTree, llvm::MachineTraceMetrics, llvm::LazyValueInfo, llvm::LiveDebugVariables, llvm::LiveRegMatrix, llvm::MachineFunctionAnalysis, llvm::MachineBlockFrequencyInfo, and llvm::BlockFrequencyInfo.

Definition at line 82 of file Pass.cpp.

                         {
  // By default, don't do anything.
}

Instruction* llvm::InstCombiner::ReplaceInstUsesWith ( Instruction &  I, Value *  V  )

inline

Definition at line 282 of file InstCombine.h.

                                                             {
    Worklist.AddUsersToWorkList(I); // Add all modified instrs to worklist.

    // If we are replacing the instruction with itself, this must be in a
    // segment of unreachable code, so just clobber the instruction.
    if (&I == V)
      V = UndefValue::get(I.getType());

    DEBUG(dbgs() << "IC: Replacing " << I << "\n"
                    "    with " << *V << '\n');

    I.replaceAllUsesWith(V);
    return &I;
  }

bool InstCombiner::runOnFunction ( Function &  F )

overridevirtual

Builder - This is an IRBuilder that automatically inserts new instructions into the worklist when they are created.

Implements llvm::FunctionPass.

Definition at line 2828 of file InstructionCombining.cpp.

                                            {
  if (skipOptnoneFunction(F))
    return false;

  DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
  DL = DLP ? &DLP->getDataLayout() : nullptr;
  TLI = &getAnalysis<TargetLibraryInfo>();
  // Minimizing size?
  MinimizeSize = F.getAttributes().hasAttribute(AttributeSet::FunctionIndex,
                                                Attribute::MinSize);

  /// Builder - This is an IRBuilder that automatically inserts new
  /// instructions into the worklist when they are created.
  IRBuilder<true, TargetFolder, InstCombineIRInserter>
    TheBuilder(F.getContext(), TargetFolder(DL),
               InstCombineIRInserter(Worklist));
  Builder = &TheBuilder;

  InstCombinerLibCallSimplifier TheSimplifier(DL, TLI, this);
  Simplifier = &TheSimplifier;

  bool EverMadeChange = false;

  // Lower dbg.declare intrinsics otherwise their value may be clobbered
  // by instcombiner.
  EverMadeChange = LowerDbgDeclare(F);

  // Iterate while there is work to do.
  unsigned Iteration = 0;
  while (DoOneIteration(F, Iteration++))
    EverMadeChange = true;

  Builder = nullptr;
  return EverMadeChange;
}

Instruction * InstCombiner::scalarizePHI ( ExtractElementInst &  EI, PHINode *  PN  )

private

Definition at line 115 of file InstCombineVectorOps.cpp.

                                                                           {
  // Verify that the PHI node has exactly 2 uses. Otherwise return NULL.
  if (!PN->hasNUses(2))
    return nullptr;

  // If so, it's known at this point that one operand is PHI and the other is
  // an extractelement node. Find the PHI user that is not the extractelement
  // node.
  auto iu = PN->user_begin();
  Instruction *PHIUser = dyn_cast<Instruction>(*iu);
  if (PHIUser == cast<Instruction>(&EI))
    PHIUser = cast<Instruction>(*(++iu));

  // Verify that this PHI user has one use, which is the PHI itself,
  // and that it is a binary operation which is cheap to scalarize.
  // otherwise return NULL.
  if (!PHIUser->hasOneUse() || !(PHIUser->user_back() == PN) ||
      !(isa<BinaryOperator>(PHIUser)) || !CheapToScalarize(PHIUser, true))
    return nullptr;

  // Create a scalar PHI node that will replace the vector PHI node
  // just before the current PHI node.
  PHINode *scalarPHI = cast<PHINode>(InsertNewInstWith(
      PHINode::Create(EI.getType(), PN->getNumIncomingValues(), ""), *PN));
  // Scalarize each PHI operand.
  for (unsigned i = 0; i < PN->getNumIncomingValues(); i++) {
    Value *PHIInVal = PN->getIncomingValue(i);
    BasicBlock *inBB = PN->getIncomingBlock(i);
    Value *Elt = EI.getIndexOperand();
    // If the operand is the PHI induction variable:
    if (PHIInVal == PHIUser) {
      // Scalarize the binary operation. Its first operand is the
      // scalar PHI, and the second operand is extracted from the other
      // vector operand.
      BinaryOperator *B0 = cast<BinaryOperator>(PHIUser);
      unsigned opId = (B0->getOperand(0) == PN) ? 1 : 0;
      Value *Op = InsertNewInstWith(
          ExtractElementInst::Create(B0->getOperand(opId), Elt,
                                     B0->getOperand(opId)->getName() + ".Elt"),
          *B0);
      Value *newPHIUser = InsertNewInstWith(
          BinaryOperator::Create(B0->getOpcode(), scalarPHI, Op), *B0);
      scalarPHI->addIncoming(newPHIUser, inBB);
    } else {
      // Scalarize PHI input:
      Instruction *newEI = ExtractElementInst::Create(PHIInVal, Elt, "");
      // Insert the new instruction into the predecessor basic block.
      Instruction *pos = dyn_cast<Instruction>(PHIInVal);
      BasicBlock::iterator InsertPos;
      if (pos && !isa<PHINode>(pos)) {
        InsertPos = pos;
        ++InsertPos;
      } else {
        InsertPos = inBB->getFirstInsertionPt();
      }

      InsertNewInstWith(newEI, *InsertPos);

      scalarPHI->addIncoming(newEI, inBB);
    }
  }
  return ReplaceInstUsesWith(EI, scalarPHI);
}

void Pass::setResolver ( AnalysisResolver *  AR )

inherited

Definition at line 102 of file Pass.cpp.

                                           {
  assert(!Resolver && "Resolver is already set");
  Resolver = AR;
}

bool InstCombiner::ShouldChangeType ( Type *  From, Type *  To  ) const

private

ShouldChangeType - Return true if it is desirable to convert a computation from 'From' to 'To'. We don't want to convert from a legal to an illegal type for example, or from a smaller to a larger illegal type.

Definition at line 105 of file InstructionCombining.cpp.

                                                              {
  assert(From->isIntegerTy() && To->isIntegerTy());

  // If we don't have DL, we don't know if the source/dest are legal.
  if (!DL) return false;

  unsigned FromWidth = From->getPrimitiveSizeInBits();
  unsigned ToWidth = To->getPrimitiveSizeInBits();
  bool FromLegal = DL->isLegalInteger(FromWidth);
  bool ToLegal = DL->isLegalInteger(ToWidth);

  // If this is a legal integer from type, and the result would be an illegal
  // type, don't do the transformation.
  if (FromLegal && !ToLegal)
    return false;

  // Otherwise, if both are illegal, do not increase the size of the result. We
  // do allow things like i160 -> i64, but not i64 -> i160.
  if (!FromLegal && !ToLegal && ToWidth > FromWidth)
    return false;

  return true;
}

bool InstCombiner::ShouldOptimizeCast ( Instruction::CastOps  opc, const Value *  V, Type *  Ty  )

private

ShouldOptimizeCast - Return true if the cast from "V to Ty" actually results in any code being generated and is interesting to optimize out. If the cast can be eliminated by some other simple transformation, we prefer to do the simplification first.

Definition at line 273 of file InstCombineCasts.cpp.

                                                {
  // Noop casts and casts of constants should be eliminated trivially.
  if (V->getType() == Ty || isa<Constant>(V)) return false;

  // If this is another cast that can be eliminated, we prefer to have it
  // eliminated.
  if (const CastInst *CI = dyn_cast<CastInst>(V))
    if (isEliminableCastPair(CI, opc, Ty, DL))
      return false;

  // If this is a vector sext from a compare, then we don't want to break the
  // idiom where each element of the extended vector is either zero or all ones.
  if (opc == Instruction::SExt && isa<CmpInst>(V) && Ty->isVectorTy())
    return false;

  return true;
}

bool InstCombiner::SimplifyAssociativeOrCommutative ( BinaryOperator &  I )

private

SimplifyAssociativeOrCommutative - This performs a few simplifications for operators which are associative or commutative.

SimplifyAssociativeOrCommutative - This performs a few simplifications for operators which are associative or commutative:

Definition at line 203 of file InstructionCombining.cpp.

                                                                     {
  Instruction::BinaryOps Opcode = I.getOpcode();
  bool Changed = false;

  do {
    // Order operands such that they are listed from right (least complex) to
    // left (most complex).  This puts constants before unary operators before
    // binary operators.
    if (I.isCommutative() && getComplexity(I.getOperand(0)) <
        getComplexity(I.getOperand(1)))
      Changed = !I.swapOperands();

    BinaryOperator *Op0 = dyn_cast<BinaryOperator>(I.getOperand(0));
    BinaryOperator *Op1 = dyn_cast<BinaryOperator>(I.getOperand(1));

    if (I.isAssociative()) {
      // Transform: "(A op B) op C" ==> "A op (B op C)" if "B op C" simplifies.
      if (Op0 && Op0->getOpcode() == Opcode) {
        Value *A = Op0->getOperand(0);
        Value *B = Op0->getOperand(1);
        Value *C = I.getOperand(1);

        // Does "B op C" simplify?
        if (Value *V = SimplifyBinOp(Opcode, B, C, DL)) {
          // It simplifies to V.  Form "A op V".
          I.setOperand(0, A);
          I.setOperand(1, V);
          // Conservatively clear the optional flags, since they may not be
          // preserved by the reassociation.
          if (MaintainNoSignedWrap(I, B, C) &&
              (!Op0 || (isa<BinaryOperator>(Op0) && Op0->hasNoSignedWrap()))) {
            // Note: this is only valid because SimplifyBinOp doesn't look at
            // the operands to Op0.
            I.clearSubclassOptionalData();
            I.setHasNoSignedWrap(true);
          } else {
            ClearSubclassDataAfterReassociation(I);
          }

          Changed = true;
          ++NumReassoc;
          continue;
        }
      }

      // Transform: "A op (B op C)" ==> "(A op B) op C" if "A op B" simplifies.
      if (Op1 && Op1->getOpcode() == Opcode) {
        Value *A = I.getOperand(0);
        Value *B = Op1->getOperand(0);
        Value *C = Op1->getOperand(1);

        // Does "A op B" simplify?
        if (Value *V = SimplifyBinOp(Opcode, A, B, DL)) {
          // It simplifies to V.  Form "V op C".
          I.setOperand(0, V);
          I.setOperand(1, C);
          // Conservatively clear the optional flags, since they may not be
          // preserved by the reassociation.
          ClearSubclassDataAfterReassociation(I);
          Changed = true;
          ++NumReassoc;
          continue;
        }
      }
    }

    if (I.isAssociative() && I.isCommutative()) {
      // Transform: "(A op B) op C" ==> "(C op A) op B" if "C op A" simplifies.
      if (Op0 && Op0->getOpcode() == Opcode) {
        Value *A = Op0->getOperand(0);
        Value *B = Op0->getOperand(1);
        Value *C = I.getOperand(1);

        // Does "C op A" simplify?
        if (Value *V = SimplifyBinOp(Opcode, C, A, DL)) {
          // It simplifies to V.  Form "V op B".
          I.setOperand(0, V);
          I.setOperand(1, B);
          // Conservatively clear the optional flags, since they may not be
          // preserved by the reassociation.
          ClearSubclassDataAfterReassociation(I);
          Changed = true;
          ++NumReassoc;
          continue;
        }
      }

      // Transform: "A op (B op C)" ==> "B op (C op A)" if "C op A" simplifies.
      if (Op1 && Op1->getOpcode() == Opcode) {
        Value *A = I.getOperand(0);
        Value *B = Op1->getOperand(0);
        Value *C = Op1->getOperand(1);

        // Does "C op A" simplify?
        if (Value *V = SimplifyBinOp(Opcode, C, A, DL)) {
          // It simplifies to V.  Form "B op V".
          I.setOperand(0, B);
          I.setOperand(1, V);
          // Conservatively clear the optional flags, since they may not be
          // preserved by the reassociation.
          ClearSubclassDataAfterReassociation(I);
          Changed = true;
          ++NumReassoc;
          continue;
        }
      }

      // Transform: "(A op C1) op (B op C2)" ==> "(A op B) op (C1 op C2)"
      // if C1 and C2 are constants.
      if (Op0 && Op1 &&
          Op0->getOpcode() == Opcode && Op1->getOpcode() == Opcode &&
          isa<Constant>(Op0->getOperand(1)) &&
          isa<Constant>(Op1->getOperand(1)) &&
          Op0->hasOneUse() && Op1->hasOneUse()) {
        Value *A = Op0->getOperand(0);
        Constant *C1 = cast<Constant>(Op0->getOperand(1));
        Value *B = Op1->getOperand(0);
        Constant *C2 = cast<Constant>(Op1->getOperand(1));

        Constant *Folded = ConstantExpr::get(Opcode, C1, C2);
        BinaryOperator *New = BinaryOperator::Create(Opcode, A, B);
        if (isa<FPMathOperator>(New)) {
          FastMathFlags Flags = I.getFastMathFlags();
          Flags &= Op0->getFastMathFlags();
          Flags &= Op1->getFastMathFlags();
          New->setFastMathFlags(Flags);
        }
        InsertNewInstWith(New, I);
        New->takeName(Op1);
        I.setOperand(0, New);
        I.setOperand(1, Folded);
        // Conservatively clear the optional flags, since they may not be
        // preserved by the reassociation.
        ClearSubclassDataAfterReassociation(I);

        Changed = true;
        continue;
      }
    }

    // No further simplifications.
    return Changed;
  } while (1);
}

bool InstCombiner::SimplifyDemandedBits ( Use &  U, APInt  DemandedMask, APInt &  KnownZero, APInt &  KnownOne, unsigned  Depth = 0  )

private

SimplifyDemandedBits - This form of SimplifyDemandedBits simplifies the specified instruction operand if possible, updating it in place. It returns true if it made any change and false otherwise.

Definition at line 70 of file InstCombineSimplifyDemanded.cpp.

                                                        {
  Value *NewVal = SimplifyDemandedUseBits(U.get(), DemandedMask,
                                          KnownZero, KnownOne, Depth);
  if (!NewVal) return false;
  U = NewVal;
  return true;
}

bool InstCombiner::SimplifyDemandedInstructionBits ( Instruction &  Inst )

private

SimplifyDemandedInstructionBits - Inst is an integer instruction that SimplifyDemandedBits knows about. See if the instruction has any properties that allow us to simplify its operands.

Definition at line 54 of file InstCombineSimplifyDemanded.cpp.

                                                                    {
  unsigned BitWidth = Inst.getType()->getScalarSizeInBits();
  APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
  APInt DemandedMask(APInt::getAllOnesValue(BitWidth));

  Value *V = SimplifyDemandedUseBits(&Inst, DemandedMask,
                                     KnownZero, KnownOne, 0);
  if (!V) return false;
  if (V == &Inst) return true;
  ReplaceInstUsesWith(Inst, V);
  return true;
}

Value * InstCombiner::SimplifyDemandedUseBits ( Value *  V, APInt  DemandedMask, APInt &  KnownZero, APInt &  KnownOne, unsigned  Depth  )

private

SimplifyDemandedUseBits - Attempts to replace V with a simpler value based on the demanded bits.

SimplifyDemandedUseBits - This function attempts to replace V with a simpler value based on the demanded bits. When this function is called, it is known that only the bits set in DemandedMask of the result of V are ever used downstream. Consequently, depending on the mask and V, it may be possible to replace V with a constant or one of its operands. In such cases, this function does the replacement and returns true. In all other cases, it returns false after analyzing the expression and setting KnownOne and known to be one in the expression. KnownZero contains all the bits that are known to be zero in the expression. These are provided to potentially allow the caller (which might recursively be SimplifyDemandedBits itself) to simplify the expression. KnownOne and KnownZero always follow the invariant that KnownOne & KnownZero == 0. That is, a bit can't be both 1 and 0. Note that the bits in KnownOne and KnownZero may only be accurate for those bits set in DemandedMask. Note also that the bitwidth of V, DemandedMask, KnownZero and KnownOne must all be the same.

This returns null if it did not change anything and it permits no simplification. This returns V itself if it did some simplification of V's operands based on the information about what bits are demanded. This returns some other non-null value if it found out that V is equal to another value in the context where the specified bits are demanded, but not for all users.

Definition at line 102 of file InstCombineSimplifyDemanded.cpp.

                                                             {
  assert(V != nullptr && "Null pointer of Value???");
  assert(Depth <= 6 && "Limit Search Depth");
  uint32_t BitWidth = DemandedMask.getBitWidth();
  Type *VTy = V->getType();
  assert((DL || !VTy->isPointerTy()) &&
         "SimplifyDemandedBits needs to know bit widths!");
  assert((!DL || DL->getTypeSizeInBits(VTy->getScalarType()) == BitWidth) &&
         (!VTy->isIntOrIntVectorTy() ||
          VTy->getScalarSizeInBits() == BitWidth) &&
         KnownZero.getBitWidth() == BitWidth &&
         KnownOne.getBitWidth() == BitWidth &&
         "Value *V, DemandedMask, KnownZero and KnownOne "
         "must have same BitWidth");
  if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
    // We know all of the bits for a constant!
    KnownOne = CI->getValue() & DemandedMask;
    KnownZero = ~KnownOne & DemandedMask;
    return nullptr;
  }
  if (isa<ConstantPointerNull>(V)) {
    // We know all of the bits for a constant!
    KnownOne.clearAllBits();
    KnownZero = DemandedMask;
    return nullptr;
  }

  KnownZero.clearAllBits();
  KnownOne.clearAllBits();
  if (DemandedMask == 0) {   // Not demanding any bits from V.
    if (isa<UndefValue>(V))
      return nullptr;
    return UndefValue::get(VTy);
  }

  if (Depth == 6)        // Limit search depth.
    return nullptr;

  APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
  APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);

  Instruction *I = dyn_cast<Instruction>(V);
  if (!I) {
    computeKnownBits(V, KnownZero, KnownOne, Depth);
    return nullptr;        // Only analyze instructions.
  }

  // If there are multiple uses of this value and we aren't at the root, then
  // we can't do any simplifications of the operands, because DemandedMask
  // only reflects the bits demanded by *one* of the users.
  if (Depth != 0 && !I->hasOneUse()) {
    // Despite the fact that we can't simplify this instruction in all User's
    // context, we can at least compute the knownzero/knownone bits, and we can
    // do simplifications that apply to *just* the one user if we know that
    // this instruction has a simpler value in that context.
    if (I->getOpcode() == Instruction::And) {
      // If either the LHS or the RHS are Zero, the result is zero.
      computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth+1);
      computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth+1);

      // If all of the demanded bits are known 1 on one side, return the other.
      // These bits cannot contribute to the result of the 'and' in this
      // context.
      if ((DemandedMask & ~LHSKnownZero & RHSKnownOne) ==
          (DemandedMask & ~LHSKnownZero))
        return I->getOperand(0);
      if ((DemandedMask & ~RHSKnownZero & LHSKnownOne) ==
          (DemandedMask & ~RHSKnownZero))
        return I->getOperand(1);

      // If all of the demanded bits in the inputs are known zeros, return zero.
      if ((DemandedMask & (RHSKnownZero|LHSKnownZero)) == DemandedMask)
        return Constant::getNullValue(VTy);

    } else if (I->getOpcode() == Instruction::Or) {
      // We can simplify (X|Y) -> X or Y in the user's context if we know that
      // only bits from X or Y are demanded.

      // If either the LHS or the RHS are One, the result is One.
      computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth+1);
      computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth+1);

      // If all of the demanded bits are known zero on one side, return the
      // other.  These bits cannot contribute to the result of the 'or' in this
      // context.
      if ((DemandedMask & ~LHSKnownOne & RHSKnownZero) ==
          (DemandedMask & ~LHSKnownOne))
        return I->getOperand(0);
      if ((DemandedMask & ~RHSKnownOne & LHSKnownZero) ==
          (DemandedMask & ~RHSKnownOne))
        return I->getOperand(1);

      // If all of the potentially set bits on one side are known to be set on
      // the other side, just use the 'other' side.
      if ((DemandedMask & (~RHSKnownZero) & LHSKnownOne) ==
          (DemandedMask & (~RHSKnownZero)))
        return I->getOperand(0);
      if ((DemandedMask & (~LHSKnownZero) & RHSKnownOne) ==
          (DemandedMask & (~LHSKnownZero)))
        return I->getOperand(1);
    } else if (I->getOpcode() == Instruction::Xor) {
      // We can simplify (X^Y) -> X or Y in the user's context if we know that
      // only bits from X or Y are demanded.

      computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth+1);
      computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth+1);

      // If all of the demanded bits are known zero on one side, return the
      // other.
      if ((DemandedMask & RHSKnownZero) == DemandedMask)
        return I->getOperand(0);
      if ((DemandedMask & LHSKnownZero) == DemandedMask)
        return I->getOperand(1);
    }

    // Compute the KnownZero/KnownOne bits to simplify things downstream.
    computeKnownBits(I, KnownZero, KnownOne, Depth);
    return nullptr;
  }

  // If this is the root being simplified, allow it to have multiple uses,
  // just set the DemandedMask to all bits so that we can try to simplify the
  // operands.  This allows visitTruncInst (for example) to simplify the
  // operand of a trunc without duplicating all the logic below.
  if (Depth == 0 && !V->hasOneUse())
    DemandedMask = APInt::getAllOnesValue(BitWidth);

  switch (I->getOpcode()) {
  default:
    computeKnownBits(I, KnownZero, KnownOne, Depth);
    break;
  case Instruction::And:
    // If either the LHS or the RHS are Zero, the result is zero.
    if (SimplifyDemandedBits(I->getOperandUse(1), DemandedMask,
                             RHSKnownZero, RHSKnownOne, Depth+1) ||
        SimplifyDemandedBits(I->getOperandUse(0), DemandedMask & ~RHSKnownZero,
                             LHSKnownZero, LHSKnownOne, Depth+1))
      return I;
    assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
    assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");

    // If all of the demanded bits are known 1 on one side, return the other.
    // These bits cannot contribute to the result of the 'and'.
    if ((DemandedMask & ~LHSKnownZero & RHSKnownOne) ==
        (DemandedMask & ~LHSKnownZero))
      return I->getOperand(0);
    if ((DemandedMask & ~RHSKnownZero & LHSKnownOne) ==
        (DemandedMask & ~RHSKnownZero))
      return I->getOperand(1);

    // If all of the demanded bits in the inputs are known zeros, return zero.
    if ((DemandedMask & (RHSKnownZero|LHSKnownZero)) == DemandedMask)
      return Constant::getNullValue(VTy);

    // If the RHS is a constant, see if we can simplify it.
    if (ShrinkDemandedConstant(I, 1, DemandedMask & ~LHSKnownZero))
      return I;

    // Output known-1 bits are only known if set in both the LHS & RHS.
    KnownOne = RHSKnownOne & LHSKnownOne;
    // Output known-0 are known to be clear if zero in either the LHS | RHS.
    KnownZero = RHSKnownZero | LHSKnownZero;
    break;
  case Instruction::Or:
    // If either the LHS or the RHS are One, the result is One.
    if (SimplifyDemandedBits(I->getOperandUse(1), DemandedMask,
                             RHSKnownZero, RHSKnownOne, Depth+1) ||
        SimplifyDemandedBits(I->getOperandUse(0), DemandedMask & ~RHSKnownOne,
                             LHSKnownZero, LHSKnownOne, Depth+1))
      return I;
    assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
    assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");

    // If all of the demanded bits are known zero on one side, return the other.
    // These bits cannot contribute to the result of the 'or'.
    if ((DemandedMask & ~LHSKnownOne & RHSKnownZero) ==
        (DemandedMask & ~LHSKnownOne))
      return I->getOperand(0);
    if ((DemandedMask & ~RHSKnownOne & LHSKnownZero) ==
        (DemandedMask & ~RHSKnownOne))
      return I->getOperand(1);

    // If all of the potentially set bits on one side are known to be set on
    // the other side, just use the 'other' side.
    if ((DemandedMask & (~RHSKnownZero) & LHSKnownOne) ==
        (DemandedMask & (~RHSKnownZero)))
      return I->getOperand(0);
    if ((DemandedMask & (~LHSKnownZero) & RHSKnownOne) ==
        (DemandedMask & (~LHSKnownZero)))
      return I->getOperand(1);

    // If the RHS is a constant, see if we can simplify it.
    if (ShrinkDemandedConstant(I, 1, DemandedMask))
      return I;

    // Output known-0 bits are only known if clear in both the LHS & RHS.
    KnownZero = RHSKnownZero & LHSKnownZero;
    // Output known-1 are known to be set if set in either the LHS | RHS.
    KnownOne = RHSKnownOne | LHSKnownOne;
    break;
  case Instruction::Xor: {
    if (SimplifyDemandedBits(I->getOperandUse(1), DemandedMask,
                             RHSKnownZero, RHSKnownOne, Depth+1) ||
        SimplifyDemandedBits(I->getOperandUse(0), DemandedMask,
                             LHSKnownZero, LHSKnownOne, Depth+1))
      return I;
    assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
    assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");

    // If all of the demanded bits are known zero on one side, return the other.
    // These bits cannot contribute to the result of the 'xor'.
    if ((DemandedMask & RHSKnownZero) == DemandedMask)
      return I->getOperand(0);
    if ((DemandedMask & LHSKnownZero) == DemandedMask)
      return I->getOperand(1);

    // If all of the demanded bits are known to be zero on one side or the
    // other, turn this into an *inclusive* or.
    //    e.g. (A & C1)^(B & C2) -> (A & C1)|(B & C2) iff C1&C2 == 0
    if ((DemandedMask & ~RHSKnownZero & ~LHSKnownZero) == 0) {
      Instruction *Or =
        BinaryOperator::CreateOr(I->getOperand(0), I->getOperand(1),
                                 I->getName());
      return InsertNewInstWith(Or, *I);
    }

    // If all of the demanded bits on one side are known, and all of the set
    // bits on that side are also known to be set on the other side, turn this
    // into an AND, as we know the bits will be cleared.
    //    e.g. (X | C1) ^ C2 --> (X | C1) & ~C2 iff (C1&C2) == C2
    if ((DemandedMask & (RHSKnownZero|RHSKnownOne)) == DemandedMask) {
      // all known
      if ((RHSKnownOne & LHSKnownOne) == RHSKnownOne) {
        Constant *AndC = Constant::getIntegerValue(VTy,
                                                   ~RHSKnownOne & DemandedMask);
        Instruction *And = BinaryOperator::CreateAnd(I->getOperand(0), AndC);
        return InsertNewInstWith(And, *I);
      }
    }

    // If the RHS is a constant, see if we can simplify it.
    // FIXME: for XOR, we prefer to force bits to 1 if they will make a -1.
    if (ShrinkDemandedConstant(I, 1, DemandedMask))
      return I;

    // If our LHS is an 'and' and if it has one use, and if any of the bits we
    // are flipping are known to be set, then the xor is just resetting those
    // bits to zero.  We can just knock out bits from the 'and' and the 'xor',
    // simplifying both of them.
    if (Instruction *LHSInst = dyn_cast<Instruction>(I->getOperand(0)))
      if (LHSInst->getOpcode() == Instruction::And && LHSInst->hasOneUse() &&
          isa<ConstantInt>(I->getOperand(1)) &&
          isa<ConstantInt>(LHSInst->getOperand(1)) &&
          (LHSKnownOne & RHSKnownOne & DemandedMask) != 0) {
        ConstantInt *AndRHS = cast<ConstantInt>(LHSInst->getOperand(1));
        ConstantInt *XorRHS = cast<ConstantInt>(I->getOperand(1));
        APInt NewMask = ~(LHSKnownOne & RHSKnownOne & DemandedMask);

        Constant *AndC =
          ConstantInt::get(I->getType(), NewMask & AndRHS->getValue());
        Instruction *NewAnd = BinaryOperator::CreateAnd(I->getOperand(0), AndC);
        InsertNewInstWith(NewAnd, *I);

        Constant *XorC =
          ConstantInt::get(I->getType(), NewMask & XorRHS->getValue());
        Instruction *NewXor = BinaryOperator::CreateXor(NewAnd, XorC);
        return InsertNewInstWith(NewXor, *I);
      }

    // Output known-0 bits are known if clear or set in both the LHS & RHS.
    KnownZero= (RHSKnownZero & LHSKnownZero) | (RHSKnownOne & LHSKnownOne);
    // Output known-1 are known to be set if set in only one of the LHS, RHS.
    KnownOne = (RHSKnownZero & LHSKnownOne) | (RHSKnownOne & LHSKnownZero);
    break;
  }
  case Instruction::Select:
    if (SimplifyDemandedBits(I->getOperandUse(2), DemandedMask,
                             RHSKnownZero, RHSKnownOne, Depth+1) ||
        SimplifyDemandedBits(I->getOperandUse(1), DemandedMask,
                             LHSKnownZero, LHSKnownOne, Depth+1))
      return I;
    assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
    assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");

    // If the operands are constants, see if we can simplify them.
    if (ShrinkDemandedConstant(I, 1, DemandedMask) ||
        ShrinkDemandedConstant(I, 2, DemandedMask))
      return I;

    // Only known if known in both the LHS and RHS.
    KnownOne = RHSKnownOne & LHSKnownOne;
    KnownZero = RHSKnownZero & LHSKnownZero;
    break;
  case Instruction::Trunc: {
    unsigned truncBf = I->getOperand(0)->getType()->getScalarSizeInBits();
    DemandedMask = DemandedMask.zext(truncBf);
    KnownZero = KnownZero.zext(truncBf);
    KnownOne = KnownOne.zext(truncBf);
    if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMask,
                             KnownZero, KnownOne, Depth+1))
      return I;
    DemandedMask = DemandedMask.trunc(BitWidth);
    KnownZero = KnownZero.trunc(BitWidth);
    KnownOne = KnownOne.trunc(BitWidth);
    assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
    break;
  }
  case Instruction::BitCast:
    if (!I->getOperand(0)->getType()->isIntOrIntVectorTy())
      return nullptr;  // vector->int or fp->int?

    if (VectorType *DstVTy = dyn_cast<VectorType>(I->getType())) {
      if (VectorType *SrcVTy =
            dyn_cast<VectorType>(I->getOperand(0)->getType())) {
        if (DstVTy->getNumElements() != SrcVTy->getNumElements())
          // Don't touch a bitcast between vectors of different element counts.
          return nullptr;
      } else
        // Don't touch a scalar-to-vector bitcast.
        return nullptr;
    } else if (I->getOperand(0)->getType()->isVectorTy())
      // Don't touch a vector-to-scalar bitcast.
      return nullptr;

    if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMask,
                             KnownZero, KnownOne, Depth+1))
      return I;
    assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
    break;
  case Instruction::ZExt: {
    // Compute the bits in the result that are not present in the input.
    unsigned SrcBitWidth =I->getOperand(0)->getType()->getScalarSizeInBits();

    DemandedMask = DemandedMask.trunc(SrcBitWidth);
    KnownZero = KnownZero.trunc(SrcBitWidth);
    KnownOne = KnownOne.trunc(SrcBitWidth);
    if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMask,
                             KnownZero, KnownOne, Depth+1))
      return I;
    DemandedMask = DemandedMask.zext(BitWidth);
    KnownZero = KnownZero.zext(BitWidth);
    KnownOne = KnownOne.zext(BitWidth);
    assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
    // The top bits are known to be zero.
    KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - SrcBitWidth);
    break;
  }
  case Instruction::SExt: {
    // Compute the bits in the result that are not present in the input.
    unsigned SrcBitWidth =I->getOperand(0)->getType()->getScalarSizeInBits();

    APInt InputDemandedBits = DemandedMask &
                              APInt::getLowBitsSet(BitWidth, SrcBitWidth);

    APInt NewBits(APInt::getHighBitsSet(BitWidth, BitWidth - SrcBitWidth));
    // If any of the sign extended bits are demanded, we know that the sign
    // bit is demanded.
    if ((NewBits & DemandedMask) != 0)
      InputDemandedBits.setBit(SrcBitWidth-1);

    InputDemandedBits = InputDemandedBits.trunc(SrcBitWidth);
    KnownZero = KnownZero.trunc(SrcBitWidth);
    KnownOne = KnownOne.trunc(SrcBitWidth);
    if (SimplifyDemandedBits(I->getOperandUse(0), InputDemandedBits,
                             KnownZero, KnownOne, Depth+1))
      return I;
    InputDemandedBits = InputDemandedBits.zext(BitWidth);
    KnownZero = KnownZero.zext(BitWidth);
    KnownOne = KnownOne.zext(BitWidth);
    assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");

    // If the sign bit of the input is known set or clear, then we know the
    // top bits of the result.

    // If the input sign bit is known zero, or if the NewBits are not demanded
    // convert this into a zero extension.
    if (KnownZero[SrcBitWidth-1] || (NewBits & ~DemandedMask) == NewBits) {
      // Convert to ZExt cast
      CastInst *NewCast = new ZExtInst(I->getOperand(0), VTy, I->getName());
      return InsertNewInstWith(NewCast, *I);
    } else if (KnownOne[SrcBitWidth-1]) {    // Input sign bit known set
      KnownOne |= NewBits;
    }
    break;
  }
  case Instruction::Add: {
    // Figure out what the input bits are.  If the top bits of the and result
    // are not demanded, then the add doesn't demand them from its input
    // either.
    unsigned NLZ = DemandedMask.countLeadingZeros();

    // If there is a constant on the RHS, there are a variety of xformations
    // we can do.
    if (ConstantInt *RHS = dyn_cast<ConstantInt>(I->getOperand(1))) {
      // If null, this should be simplified elsewhere.  Some of the xforms here
      // won't work if the RHS is zero.
      if (RHS->isZero())
        break;

      // If the top bit of the output is demanded, demand everything from the
      // input.  Otherwise, we demand all the input bits except NLZ top bits.
      APInt InDemandedBits(APInt::getLowBitsSet(BitWidth, BitWidth - NLZ));

      // Find information about known zero/one bits in the input.
      if (SimplifyDemandedBits(I->getOperandUse(0), InDemandedBits,
                               LHSKnownZero, LHSKnownOne, Depth+1))
        return I;

      // If the RHS of the add has bits set that can't affect the input, reduce
      // the constant.
      if (ShrinkDemandedConstant(I, 1, InDemandedBits))
        return I;

      // Avoid excess work.
      if (LHSKnownZero == 0 && LHSKnownOne == 0)
        break;

      // Turn it into OR if input bits are zero.
      if ((LHSKnownZero & RHS->getValue()) == RHS->getValue()) {
        Instruction *Or =
          BinaryOperator::CreateOr(I->getOperand(0), I->getOperand(1),
                                   I->getName());
        return InsertNewInstWith(Or, *I);
      }

      // We can say something about the output known-zero and known-one bits,
      // depending on potential carries from the input constant and the
      // unknowns.  For example if the LHS is known to have at most the 0x0F0F0
      // bits set and the RHS constant is 0x01001, then we know we have a known
      // one mask of 0x00001 and a known zero mask of 0xE0F0E.

      // To compute this, we first compute the potential carry bits.  These are
      // the bits which may be modified.  I'm not aware of a better way to do
      // this scan.
      const APInt &RHSVal = RHS->getValue();
      APInt CarryBits((~LHSKnownZero + RHSVal) ^ (~LHSKnownZero ^ RHSVal));

      // Now that we know which bits have carries, compute the known-1/0 sets.

      // Bits are known one if they are known zero in one operand and one in the
      // other, and there is no input carry.
      KnownOne = ((LHSKnownZero & RHSVal) |
                  (LHSKnownOne & ~RHSVal)) & ~CarryBits;

      // Bits are known zero if they are known zero in both operands and there
      // is no input carry.
      KnownZero = LHSKnownZero & ~RHSVal & ~CarryBits;
    } else {
      // If the high-bits of this ADD are not demanded, then it does not demand
      // the high bits of its LHS or RHS.
      if (DemandedMask[BitWidth-1] == 0) {
        // Right fill the mask of bits for this ADD to demand the most
        // significant bit and all those below it.
        APInt DemandedFromOps(APInt::getLowBitsSet(BitWidth, BitWidth-NLZ));
        if (SimplifyDemandedBits(I->getOperandUse(0), DemandedFromOps,
                                 LHSKnownZero, LHSKnownOne, Depth+1) ||
            SimplifyDemandedBits(I->getOperandUse(1), DemandedFromOps,
                                 LHSKnownZero, LHSKnownOne, Depth+1))
          return I;
      }
    }
    break;
  }
  case Instruction::Sub:
    // If the high-bits of this SUB are not demanded, then it does not demand
    // the high bits of its LHS or RHS.
    if (DemandedMask[BitWidth-1] == 0) {
      // Right fill the mask of bits for this SUB to demand the most
      // significant bit and all those below it.
      uint32_t NLZ = DemandedMask.countLeadingZeros();
      APInt DemandedFromOps(APInt::getLowBitsSet(BitWidth, BitWidth-NLZ));
      if (SimplifyDemandedBits(I->getOperandUse(0), DemandedFromOps,
                               LHSKnownZero, LHSKnownOne, Depth+1) ||
          SimplifyDemandedBits(I->getOperandUse(1), DemandedFromOps,
                               LHSKnownZero, LHSKnownOne, Depth+1))
        return I;
    }

    // Otherwise just hand the sub off to computeKnownBits to fill in
    // the known zeros and ones.
    computeKnownBits(V, KnownZero, KnownOne, Depth);

    // Turn this into a xor if LHS is 2^n-1 and the remaining bits are known
    // zero.
    if (ConstantInt *C0 = dyn_cast<ConstantInt>(I->getOperand(0))) {
      APInt I0 = C0->getValue();
      if ((I0 + 1).isPowerOf2() && (I0 | KnownZero).isAllOnesValue()) {
        Instruction *Xor = BinaryOperator::CreateXor(I->getOperand(1), C0);
        return InsertNewInstWith(Xor, *I);
      }
    }
    break;
  case Instruction::Shl:
    if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
      {
        Value *VarX; ConstantInt *C1;
        if (match(I->getOperand(0), m_Shr(m_Value(VarX), m_ConstantInt(C1)))) {
          Instruction *Shr = cast<Instruction>(I->getOperand(0));
          Value *R = SimplifyShrShlDemandedBits(Shr, I, DemandedMask,
                                                KnownZero, KnownOne);
          if (R)
            return R;
        }
      }

      uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
      APInt DemandedMaskIn(DemandedMask.lshr(ShiftAmt));

      // If the shift is NUW/NSW, then it does demand the high bits.
      ShlOperator *IOp = cast<ShlOperator>(I);
      if (IOp->hasNoSignedWrap())
        DemandedMaskIn |= APInt::getHighBitsSet(BitWidth, ShiftAmt+1);
      else if (IOp->hasNoUnsignedWrap())
        DemandedMaskIn |= APInt::getHighBitsSet(BitWidth, ShiftAmt);

      if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMaskIn,
                               KnownZero, KnownOne, Depth+1))
        return I;
      assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
      KnownZero <<= ShiftAmt;
      KnownOne  <<= ShiftAmt;
      // low bits known zero.
      if (ShiftAmt)
        KnownZero |= APInt::getLowBitsSet(BitWidth, ShiftAmt);
    }
    break;
  case Instruction::LShr:
    // For a logical shift right
    if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
      uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1);

      // Unsigned shift right.
      APInt DemandedMaskIn(DemandedMask.shl(ShiftAmt));

      // If the shift is exact, then it does demand the low bits (and knows that
      // they are zero).
      if (cast<LShrOperator>(I)->isExact())
        DemandedMaskIn |= APInt::getLowBitsSet(BitWidth, ShiftAmt);

      if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMaskIn,
                               KnownZero, KnownOne, Depth+1))
        return I;
      assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
      KnownZero = APIntOps::lshr(KnownZero, ShiftAmt);
      KnownOne  = APIntOps::lshr(KnownOne, ShiftAmt);
      if (ShiftAmt) {
        // Compute the new bits that are at the top now.
        APInt HighBits(APInt::getHighBitsSet(BitWidth, ShiftAmt));
        KnownZero |= HighBits;  // high bits known zero.
      }
    }
    break;
  case Instruction::AShr:
    // If this is an arithmetic shift right and only the low-bit is set, we can
    // always convert this into a logical shr, even if the shift amount is
    // variable.  The low bit of the shift cannot be an input sign bit unless
    // the shift amount is >= the size of the datatype, which is undefined.
    if (DemandedMask == 1) {
      // Perform the logical shift right.
      Instruction *NewVal = BinaryOperator::CreateLShr(
                        I->getOperand(0), I->getOperand(1), I->getName());
      return InsertNewInstWith(NewVal, *I);
    }

    // If the sign bit is the only bit demanded by this ashr, then there is no
    // need to do it, the shift doesn't change the high bit.
    if (DemandedMask.isSignBit())
      return I->getOperand(0);

    if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
      uint32_t ShiftAmt = SA->getLimitedValue(BitWidth-1);

      // Signed shift right.
      APInt DemandedMaskIn(DemandedMask.shl(ShiftAmt));
      // If any of the "high bits" are demanded, we should set the sign bit as
      // demanded.
      if (DemandedMask.countLeadingZeros() <= ShiftAmt)
        DemandedMaskIn.setBit(BitWidth-1);

      // If the shift is exact, then it does demand the low bits (and knows that
      // they are zero).
      if (cast<AShrOperator>(I)->isExact())
        DemandedMaskIn |= APInt::getLowBitsSet(BitWidth, ShiftAmt);

      if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMaskIn,
                               KnownZero, KnownOne, Depth+1))
        return I;
      assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
      // Compute the new bits that are at the top now.
      APInt HighBits(APInt::getHighBitsSet(BitWidth, ShiftAmt));
      KnownZero = APIntOps::lshr(KnownZero, ShiftAmt);
      KnownOne  = APIntOps::lshr(KnownOne, ShiftAmt);

      // Handle the sign bits.
      APInt SignBit(APInt::getSignBit(BitWidth));
      // Adjust to where it is now in the mask.
      SignBit = APIntOps::lshr(SignBit, ShiftAmt);

      // If the input sign bit is known to be zero, or if none of the top bits
      // are demanded, turn this into an unsigned shift right.
      if (BitWidth <= ShiftAmt || KnownZero[BitWidth-ShiftAmt-1] ||
          (HighBits & ~DemandedMask) == HighBits) {
        // Perform the logical shift right.
        BinaryOperator *NewVal = BinaryOperator::CreateLShr(I->getOperand(0),
                                                            SA, I->getName());
        NewVal->setIsExact(cast<BinaryOperator>(I)->isExact());
        return InsertNewInstWith(NewVal, *I);
      } else if ((KnownOne & SignBit) != 0) { // New bits are known one.
        KnownOne |= HighBits;
      }
    }
    break;
  case Instruction::SRem:
    if (ConstantInt *Rem = dyn_cast<ConstantInt>(I->getOperand(1))) {
      // X % -1 demands all the bits because we don't want to introduce
      // INT_MIN % -1 (== undef) by accident.
      if (Rem->isAllOnesValue())
        break;
      APInt RA = Rem->getValue().abs();
      if (RA.isPowerOf2()) {
        if (DemandedMask.ult(RA))    // srem won't affect demanded bits
          return I->getOperand(0);

        APInt LowBits = RA - 1;
        APInt Mask2 = LowBits | APInt::getSignBit(BitWidth);
        if (SimplifyDemandedBits(I->getOperandUse(0), Mask2,
                                 LHSKnownZero, LHSKnownOne, Depth+1))
          return I;

        // The low bits of LHS are unchanged by the srem.
        KnownZero = LHSKnownZero & LowBits;
        KnownOne = LHSKnownOne & LowBits;

        // If LHS is non-negative or has all low bits zero, then the upper bits
        // are all zero.
        if (LHSKnownZero[BitWidth-1] || ((LHSKnownZero & LowBits) == LowBits))
          KnownZero |= ~LowBits;

        // If LHS is negative and not all low bits are zero, then the upper bits
        // are all one.
        if (LHSKnownOne[BitWidth-1] && ((LHSKnownOne & LowBits) != 0))
          KnownOne |= ~LowBits;

        assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
      }
    }

    // The sign bit is the LHS's sign bit, except when the result of the
    // remainder is zero.
    if (DemandedMask.isNegative() && KnownZero.isNonNegative()) {
      APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
      computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth+1);
      // If it's known zero, our sign bit is also zero.
      if (LHSKnownZero.isNegative())
        KnownZero.setBit(KnownZero.getBitWidth() - 1);
    }
    break;
  case Instruction::URem: {
    APInt KnownZero2(BitWidth, 0), KnownOne2(BitWidth, 0);
    APInt AllOnes = APInt::getAllOnesValue(BitWidth);
    if (SimplifyDemandedBits(I->getOperandUse(0), AllOnes,
                             KnownZero2, KnownOne2, Depth+1) ||
        SimplifyDemandedBits(I->getOperandUse(1), AllOnes,
                             KnownZero2, KnownOne2, Depth+1))
      return I;

    unsigned Leaders = KnownZero2.countLeadingOnes();
    Leaders = std::max(Leaders,
                       KnownZero2.countLeadingOnes());
    KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & DemandedMask;
    break;
  }
  case Instruction::Call:
    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
      switch (II->getIntrinsicID()) {
      default: break;
      case Intrinsic::bswap: {
        // If the only bits demanded come from one byte of the bswap result,
        // just shift the input byte into position to eliminate the bswap.
        unsigned NLZ = DemandedMask.countLeadingZeros();
        unsigned NTZ = DemandedMask.countTrailingZeros();

        // Round NTZ down to the next byte.  If we have 11 trailing zeros, then
        // we need all the bits down to bit 8.  Likewise, round NLZ.  If we
        // have 14 leading zeros, round to 8.
        NLZ &= ~7;
        NTZ &= ~7;
        // If we need exactly one byte, we can do this transformation.
        if (BitWidth-NLZ-NTZ == 8) {
          unsigned ResultBit = NTZ;
          unsigned InputBit = BitWidth-NTZ-8;

          // Replace this with either a left or right shift to get the byte into
          // the right place.
          Instruction *NewVal;
          if (InputBit > ResultBit)
            NewVal = BinaryOperator::CreateLShr(II->getArgOperand(0),
                    ConstantInt::get(I->getType(), InputBit-ResultBit));
          else
            NewVal = BinaryOperator::CreateShl(II->getArgOperand(0),
                    ConstantInt::get(I->getType(), ResultBit-InputBit));
          NewVal->takeName(I);
          return InsertNewInstWith(NewVal, *I);
        }

        // TODO: Could compute known zero/one bits based on the input.
        break;
      }
      case Intrinsic::x86_sse42_crc32_64_64:
        KnownZero = APInt::getHighBitsSet(64, 32);
        return nullptr;
      }
    }
    computeKnownBits(V, KnownZero, KnownOne, Depth);
    break;
  }

  // If the client is only demanding bits that we know, return the known
  // constant.
  if ((DemandedMask & (KnownZero|KnownOne)) == DemandedMask)
    return Constant::getIntegerValue(VTy, KnownOne);
  return nullptr;
}

Value * InstCombiner::SimplifyDemandedVectorElts ( Value *  V, APInt  DemandedElts, APInt &  UndefElts, unsigned  Depth = 0  )

private

SimplifyDemandedVectorElts - The specified value produces a vector with any number of elements. DemandedElts contains the set of elements that are actually used by the caller. This method analyzes which elements of the operand are undef and returns that information in UndefElts.

If the information about demanded elements can be used to simplify the operation, the operation is simplified, then the resultant value is returned. This returns null if no change was made.

Definition at line 917 of file InstCombineSimplifyDemanded.cpp.

                                                                {
  unsigned VWidth = cast<VectorType>(V->getType())->getNumElements();
  APInt EltMask(APInt::getAllOnesValue(VWidth));
  assert((DemandedElts & ~EltMask) == 0 && "Invalid DemandedElts!");

  if (isa<UndefValue>(V)) {
    // If the entire vector is undefined, just return this info.
    UndefElts = EltMask;
    return nullptr;
  }

  if (DemandedElts == 0) { // If nothing is demanded, provide undef.
    UndefElts = EltMask;
    return UndefValue::get(V->getType());
  }

  UndefElts = 0;

  // Handle ConstantAggregateZero, ConstantVector, ConstantDataSequential.
  if (Constant *C = dyn_cast<Constant>(V)) {
    // Check if this is identity. If so, return 0 since we are not simplifying
    // anything.
    if (DemandedElts.isAllOnesValue())
      return nullptr;

    Type *EltTy = cast<VectorType>(V->getType())->getElementType();
    Constant *Undef = UndefValue::get(EltTy);

    SmallVector<Constant*, 16> Elts;
    for (unsigned i = 0; i != VWidth; ++i) {
      if (!DemandedElts[i]) {   // If not demanded, set to undef.
        Elts.push_back(Undef);
        UndefElts.setBit(i);
        continue;
      }

      Constant *Elt = C->getAggregateElement(i);
      if (!Elt) return nullptr;

      if (isa<UndefValue>(Elt)) {   // Already undef.
        Elts.push_back(Undef);
        UndefElts.setBit(i);
      } else {                               // Otherwise, defined.
        Elts.push_back(Elt);
      }
    }

    // If we changed the constant, return it.
    Constant *NewCV = ConstantVector::get(Elts);
    return NewCV != C ? NewCV : nullptr;
  }

  // Limit search depth.
  if (Depth == 10)
    return nullptr;

  // If multiple users are using the root value, proceed with
  // simplification conservatively assuming that all elements
  // are needed.
  if (!V->hasOneUse()) {
    // Quit if we find multiple users of a non-root value though.
    // They'll be handled when it's their turn to be visited by
    // the main instcombine process.
    if (Depth != 0)
      // TODO: Just compute the UndefElts information recursively.
      return nullptr;

    // Conservatively assume that all elements are needed.
    DemandedElts = EltMask;
  }

  Instruction *I = dyn_cast<Instruction>(V);
  if (!I) return nullptr;        // Only analyze instructions.

  bool MadeChange = false;
  APInt UndefElts2(VWidth, 0);
  Value *TmpV;
  switch (I->getOpcode()) {
  default: break;

  case Instruction::InsertElement: {
    // If this is a variable index, we don't know which element it overwrites.
    // demand exactly the same input as we produce.
    ConstantInt *Idx = dyn_cast<ConstantInt>(I->getOperand(2));
    if (!Idx) {
      // Note that we can't propagate undef elt info, because we don't know
      // which elt is getting updated.
      TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts,
                                        UndefElts2, Depth+1);
      if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
      break;
    }

    // If this is inserting an element that isn't demanded, remove this
    // insertelement.
    unsigned IdxNo = Idx->getZExtValue();
    if (IdxNo >= VWidth || !DemandedElts[IdxNo]) {
      Worklist.Add(I);
      return I->getOperand(0);
    }

    // Otherwise, the element inserted overwrites whatever was there, so the
    // input demanded set is simpler than the output set.
    APInt DemandedElts2 = DemandedElts;
    DemandedElts2.clearBit(IdxNo);
    TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts2,
                                      UndefElts, Depth+1);
    if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }

    // The inserted element is defined.
    UndefElts.clearBit(IdxNo);
    break;
  }
  case Instruction::ShuffleVector: {
    ShuffleVectorInst *Shuffle = cast<ShuffleVectorInst>(I);
    uint64_t LHSVWidth =
      cast<VectorType>(Shuffle->getOperand(0)->getType())->getNumElements();
    APInt LeftDemanded(LHSVWidth, 0), RightDemanded(LHSVWidth, 0);
    for (unsigned i = 0; i < VWidth; i++) {
      if (DemandedElts[i]) {
        unsigned MaskVal = Shuffle->getMaskValue(i);
        if (MaskVal != -1u) {
          assert(MaskVal < LHSVWidth * 2 &&
                 "shufflevector mask index out of range!");
          if (MaskVal < LHSVWidth)
            LeftDemanded.setBit(MaskVal);
          else
            RightDemanded.setBit(MaskVal - LHSVWidth);
        }
      }
    }

    APInt UndefElts4(LHSVWidth, 0);
    TmpV = SimplifyDemandedVectorElts(I->getOperand(0), LeftDemanded,
                                      UndefElts4, Depth+1);
    if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }

    APInt UndefElts3(LHSVWidth, 0);
    TmpV = SimplifyDemandedVectorElts(I->getOperand(1), RightDemanded,
                                      UndefElts3, Depth+1);
    if (TmpV) { I->setOperand(1, TmpV); MadeChange = true; }

    bool NewUndefElts = false;
    for (unsigned i = 0; i < VWidth; i++) {
      unsigned MaskVal = Shuffle->getMaskValue(i);
      if (MaskVal == -1u) {
        UndefElts.setBit(i);
      } else if (!DemandedElts[i]) {
        NewUndefElts = true;
        UndefElts.setBit(i);
      } else if (MaskVal < LHSVWidth) {
        if (UndefElts4[MaskVal]) {
          NewUndefElts = true;
          UndefElts.setBit(i);
        }
      } else {
        if (UndefElts3[MaskVal - LHSVWidth]) {
          NewUndefElts = true;
          UndefElts.setBit(i);
        }
      }
    }

    if (NewUndefElts) {
      // Add additional discovered undefs.
      SmallVector<Constant*, 16> Elts;
      for (unsigned i = 0; i < VWidth; ++i) {
        if (UndefElts[i])
          Elts.push_back(UndefValue::get(Type::getInt32Ty(I->getContext())));
        else
          Elts.push_back(ConstantInt::get(Type::getInt32Ty(I->getContext()),
                                          Shuffle->getMaskValue(i)));
      }
      I->setOperand(2, ConstantVector::get(Elts));
      MadeChange = true;
    }
    break;
  }
  case Instruction::Select: {
    APInt LeftDemanded(DemandedElts), RightDemanded(DemandedElts);
    if (ConstantVector* CV = dyn_cast<ConstantVector>(I->getOperand(0))) {
      for (unsigned i = 0; i < VWidth; i++) {
        if (CV->getAggregateElement(i)->isNullValue())
          LeftDemanded.clearBit(i);
        else
          RightDemanded.clearBit(i);
      }
    }

    TmpV = SimplifyDemandedVectorElts(I->getOperand(1), LeftDemanded,
                                      UndefElts, Depth+1);
    if (TmpV) { I->setOperand(1, TmpV); MadeChange = true; }

    TmpV = SimplifyDemandedVectorElts(I->getOperand(2), RightDemanded,
                                      UndefElts2, Depth+1);
    if (TmpV) { I->setOperand(2, TmpV); MadeChange = true; }

    // Output elements are undefined if both are undefined.
    UndefElts &= UndefElts2;
    break;
  }
  case Instruction::BitCast: {
    // Vector->vector casts only.
    VectorType *VTy = dyn_cast<VectorType>(I->getOperand(0)->getType());
    if (!VTy) break;
    unsigned InVWidth = VTy->getNumElements();
    APInt InputDemandedElts(InVWidth, 0);
    unsigned Ratio;

    if (VWidth == InVWidth) {
      // If we are converting from <4 x i32> -> <4 x f32>, we demand the same
      // elements as are demanded of us.
      Ratio = 1;
      InputDemandedElts = DemandedElts;
    } else if (VWidth > InVWidth) {
      // Untested so far.
      break;

      // If there are more elements in the result than there are in the source,
      // then an input element is live if any of the corresponding output
      // elements are live.
      Ratio = VWidth/InVWidth;
      for (unsigned OutIdx = 0; OutIdx != VWidth; ++OutIdx) {
        if (DemandedElts[OutIdx])
          InputDemandedElts.setBit(OutIdx/Ratio);
      }
    } else {
      // Untested so far.
      break;

      // If there are more elements in the source than there are in the result,
      // then an input element is live if the corresponding output element is
      // live.
      Ratio = InVWidth/VWidth;
      for (unsigned InIdx = 0; InIdx != InVWidth; ++InIdx)
        if (DemandedElts[InIdx/Ratio])
          InputDemandedElts.setBit(InIdx);
    }

    // div/rem demand all inputs, because they don't want divide by zero.
    TmpV = SimplifyDemandedVectorElts(I->getOperand(0), InputDemandedElts,
                                      UndefElts2, Depth+1);
    if (TmpV) {
      I->setOperand(0, TmpV);
      MadeChange = true;
    }

    UndefElts = UndefElts2;
    if (VWidth > InVWidth) {
      llvm_unreachable("Unimp");
      // If there are more elements in the result than there are in the source,
      // then an output element is undef if the corresponding input element is
      // undef.
      for (unsigned OutIdx = 0; OutIdx != VWidth; ++OutIdx)
        if (UndefElts2[OutIdx/Ratio])
          UndefElts.setBit(OutIdx);
    } else if (VWidth < InVWidth) {
      llvm_unreachable("Unimp");
      // If there are more elements in the source than there are in the result,
      // then a result element is undef if all of the corresponding input
      // elements are undef.
      UndefElts = ~0ULL >> (64-VWidth);  // Start out all undef.
      for (unsigned InIdx = 0; InIdx != InVWidth; ++InIdx)
        if (!UndefElts2[InIdx])            // Not undef?
          UndefElts.clearBit(InIdx/Ratio);    // Clear undef bit.
    }
    break;
  }
  case Instruction::And:
  case Instruction::Or:
  case Instruction::Xor:
  case Instruction::Add:
  case Instruction::Sub:
  case Instruction::Mul:
    // div/rem demand all inputs, because they don't want divide by zero.
    TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts,
                                      UndefElts, Depth+1);
    if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
    TmpV = SimplifyDemandedVectorElts(I->getOperand(1), DemandedElts,
                                      UndefElts2, Depth+1);
    if (TmpV) { I->setOperand(1, TmpV); MadeChange = true; }

    // Output elements are undefined if both are undefined.  Consider things
    // like undef&0.  The result is known zero, not undef.
    UndefElts &= UndefElts2;
    break;
  case Instruction::FPTrunc:
  case Instruction::FPExt:
    TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts,
                                      UndefElts, Depth+1);
    if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
    break;

  case Instruction::Call: {
    IntrinsicInst *II = dyn_cast<IntrinsicInst>(I);
    if (!II) break;
    switch (II->getIntrinsicID()) {
    default: break;

    // Binary vector operations that work column-wise.  A dest element is a
    // function of the corresponding input elements from the two inputs.
    case Intrinsic::x86_sse_sub_ss:
    case Intrinsic::x86_sse_mul_ss:
    case Intrinsic::x86_sse_min_ss:
    case Intrinsic::x86_sse_max_ss:
    case Intrinsic::x86_sse2_sub_sd:
    case Intrinsic::x86_sse2_mul_sd:
    case Intrinsic::x86_sse2_min_sd:
    case Intrinsic::x86_sse2_max_sd:
      TmpV = SimplifyDemandedVectorElts(II->getArgOperand(0), DemandedElts,
                                        UndefElts, Depth+1);
      if (TmpV) { II->setArgOperand(0, TmpV); MadeChange = true; }
      TmpV = SimplifyDemandedVectorElts(II->getArgOperand(1), DemandedElts,
                                        UndefElts2, Depth+1);
      if (TmpV) { II->setArgOperand(1, TmpV); MadeChange = true; }

      // If only the low elt is demanded and this is a scalarizable intrinsic,
      // scalarize it now.
      if (DemandedElts == 1) {
        switch (II->getIntrinsicID()) {
        default: break;
        case Intrinsic::x86_sse_sub_ss:
        case Intrinsic::x86_sse_mul_ss:
        case Intrinsic::x86_sse2_sub_sd:
        case Intrinsic::x86_sse2_mul_sd:
          // TODO: Lower MIN/MAX/ABS/etc
          Value *LHS = II->getArgOperand(0);
          Value *RHS = II->getArgOperand(1);
          // Extract the element as scalars.
          LHS = InsertNewInstWith(ExtractElementInst::Create(LHS,
            ConstantInt::get(Type::getInt32Ty(I->getContext()), 0U)), *II);
          RHS = InsertNewInstWith(ExtractElementInst::Create(RHS,
            ConstantInt::get(Type::getInt32Ty(I->getContext()), 0U)), *II);

          switch (II->getIntrinsicID()) {
          default: llvm_unreachable("Case stmts out of sync!");
          case Intrinsic::x86_sse_sub_ss:
          case Intrinsic::x86_sse2_sub_sd:
            TmpV = InsertNewInstWith(BinaryOperator::CreateFSub(LHS, RHS,
                                                        II->getName()), *II);
            break;
          case Intrinsic::x86_sse_mul_ss:
          case Intrinsic::x86_sse2_mul_sd:
            TmpV = InsertNewInstWith(BinaryOperator::CreateFMul(LHS, RHS,
                                                         II->getName()), *II);
            break;
          }

          Instruction *New =
            InsertElementInst::Create(
              UndefValue::get(II->getType()), TmpV,
              ConstantInt::get(Type::getInt32Ty(I->getContext()), 0U, false),
                                      II->getName());
          InsertNewInstWith(New, *II);
          return New;
        }
      }

      // Output elements are undefined if both are undefined.  Consider things
      // like undef&0.  The result is known zero, not undef.
      UndefElts &= UndefElts2;
      break;
    }
    break;
  }
  }
  return MadeChange ? I : nullptr;
}

bool InstCombiner::SimplifyDivRemOfSelect

(

BinaryOperator &

I

)

SimplifyDivRemOfSelect - Try to fold a divide or remainder of a select instruction.

Definition at line 640 of file InstCombineMulDivRem.cpp.

                                                           {
  SelectInst *SI = cast<SelectInst>(I.getOperand(1));

  // div/rem X, (Cond ? 0 : Y) -> div/rem X, Y
  int NonNullOperand = -1;
  if (Constant *ST = dyn_cast<Constant>(SI->getOperand(1)))
    if (ST->isNullValue())
      NonNullOperand = 2;
  // div/rem X, (Cond ? Y : 0) -> div/rem X, Y
  if (Constant *ST = dyn_cast<Constant>(SI->getOperand(2)))
    if (ST->isNullValue())
      NonNullOperand = 1;

  if (NonNullOperand == -1)
    return false;

  Value *SelectCond = SI->getOperand(0);

  // Change the div/rem to use 'Y' instead of the select.
  I.setOperand(1, SI->getOperand(NonNullOperand));

  // Okay, we know we replace the operand of the div/rem with 'Y' with no
  // problem.  However, the select, or the condition of the select may have
  // multiple uses.  Based on our knowledge that the operand must be non-zero,
  // propagate the known value for the select into other uses of it, and
  // propagate a known value of the condition into its other users.

  // If the select and condition only have a single use, don't bother with this,
  // early exit.
  if (SI->use_empty() && SelectCond->hasOneUse())
    return true;

  // Scan the current block backward, looking for other uses of SI.
  BasicBlock::iterator BBI = &I, BBFront = I.getParent()->begin();

  while (BBI != BBFront) {
    --BBI;
    // If we found a call to a function, we can't assume it will return, so
    // information from below it cannot be propagated above it.
    if (isa<CallInst>(BBI) && !isa<IntrinsicInst>(BBI))
      break;

    // Replace uses of the select or its condition with the known values.
    for (Instruction::op_iterator I = BBI->op_begin(), E = BBI->op_end();
         I != E; ++I) {
      if (*I == SI) {
        *I = SI->getOperand(NonNullOperand);
        Worklist.Add(BBI);
      } else if (*I == SelectCond) {
        *I = Builder->getInt1(NonNullOperand == 1);
        Worklist.Add(BBI);
      }
    }

    // If we past the instruction, quit looking for it.
    if (&*BBI == SI)
      SI = nullptr;
    if (&*BBI == SelectCond)
      SelectCond = nullptr;

    // If we ran out of things to eliminate, break out of the loop.
    if (!SelectCond && !SI)
      break;

  }
  return true;
}

Instruction * InstCombiner::SimplifyMemSet ( MemSetInst *  MI )

private

Definition at line 156 of file InstCombineCalls.cpp.

                                                        {
  unsigned Alignment = getKnownAlignment(MI->getDest(), DL);
  if (MI->getAlignment() < Alignment) {
    MI->setAlignment(ConstantInt::get(MI->getAlignmentType(),
                                             Alignment, false));
    return MI;
  }

  // Extract the length and alignment and fill if they are constant.
  ConstantInt *LenC = dyn_cast<ConstantInt>(MI->getLength());
  ConstantInt *FillC = dyn_cast<ConstantInt>(MI->getValue());
  if (!LenC || !FillC || !FillC->getType()->isIntegerTy(8))
    return nullptr;
  uint64_t Len = LenC->getLimitedValue();
  Alignment = MI->getAlignment();
  assert(Len && "0-sized memory setting should be removed already.");

  // memset(s,c,n) -> store s, c (for n=1,2,4,8)
  if (Len <= 8 && isPowerOf2_32((uint32_t)Len)) {
    Type *ITy = IntegerType::get(MI->getContext(), Len*8);  // n=1 -> i8.

    Value *Dest = MI->getDest();
    unsigned DstAddrSp = cast<PointerType>(Dest->getType())->getAddressSpace();
    Type *NewDstPtrTy = PointerType::get(ITy, DstAddrSp);
    Dest = Builder->CreateBitCast(Dest, NewDstPtrTy);

    // Alignment 0 is identity for alignment 1 for memset, but not store.
    if (Alignment == 0) Alignment = 1;

    // Extract the fill value and store.
    uint64_t Fill = FillC->getZExtValue()*0x0101010101010101ULL;
    StoreInst *S = Builder->CreateStore(ConstantInt::get(ITy, Fill), Dest,
                                        MI->isVolatile());
    S->setAlignment(Alignment);

    // Set the size of the copy to 0, it will be deleted on the next iteration.
    MI->setLength(Constant::getNullValue(LenC->getType()));
    return MI;
  }

  return nullptr;
}

Instruction * InstCombiner::SimplifyMemTransfer ( MemIntrinsic *  MI )

private

Definition at line 60 of file InstCombineCalls.cpp.

                                                               {
  unsigned DstAlign = getKnownAlignment(MI->getArgOperand(0), DL);
  unsigned SrcAlign = getKnownAlignment(MI->getArgOperand(1), DL);
  unsigned MinAlign = std::min(DstAlign, SrcAlign);
  unsigned CopyAlign = MI->getAlignment();

  if (CopyAlign < MinAlign) {
    MI->setAlignment(ConstantInt::get(MI->getAlignmentType(),
                                             MinAlign, false));
    return MI;
  }

  // If MemCpyInst length is 1/2/4/8 bytes then replace memcpy with
  // load/store.
  ConstantInt *MemOpLength = dyn_cast<ConstantInt>(MI->getArgOperand(2));
  if (!MemOpLength) return nullptr;

  // Source and destination pointer types are always "i8*" for intrinsic.  See
  // if the size is something we can handle with a single primitive load/store.
  // A single load+store correctly handles overlapping memory in the memmove
  // case.
  uint64_t Size = MemOpLength->getLimitedValue();
  assert(Size && "0-sized memory transferring should be removed already.");

  if (Size > 8 || (Size&(Size-1)))
    return nullptr;  // If not 1/2/4/8 bytes, exit.

  // Use an integer load+store unless we can find something better.
  unsigned SrcAddrSp =
    cast<PointerType>(MI->getArgOperand(1)->getType())->getAddressSpace();
  unsigned DstAddrSp =
    cast<PointerType>(MI->getArgOperand(0)->getType())->getAddressSpace();

  IntegerType* IntType = IntegerType::get(MI->getContext(), Size<<3);
  Type *NewSrcPtrTy = PointerType::get(IntType, SrcAddrSp);
  Type *NewDstPtrTy = PointerType::get(IntType, DstAddrSp);

  // Memcpy forces the use of i8* for the source and destination.  That means
  // that if you're using memcpy to move one double around, you'll get a cast
  // from double* to i8*.  We'd much rather use a double load+store rather than
  // an i64 load+store, here because this improves the odds that the source or
  // dest address will be promotable.  See if we can find a better type than the
  // integer datatype.
  Value *StrippedDest = MI->getArgOperand(0)->stripPointerCasts();
  MDNode *CopyMD = nullptr;
  if (StrippedDest != MI->getArgOperand(0)) {
    Type *SrcETy = cast<PointerType>(StrippedDest->getType())
                                    ->getElementType();
    if (DL && SrcETy->isSized() && DL->getTypeStoreSize(SrcETy) == Size) {
      // The SrcETy might be something like {{{double}}} or [1 x double].  Rip
      // down through these levels if so.
      SrcETy = reduceToSingleValueType(SrcETy);

      if (SrcETy->isSingleValueType()) {
        NewSrcPtrTy = PointerType::get(SrcETy, SrcAddrSp);
        NewDstPtrTy = PointerType::get(SrcETy, DstAddrSp);

        // If the memcpy has metadata describing the members, see if we can
        // get the TBAA tag describing our copy.
        if (MDNode *M = MI->getMetadata(LLVMContext::MD_tbaa_struct)) {
          if (M->getNumOperands() == 3 &&
              M->getOperand(0) &&
              isa<ConstantInt>(M->getOperand(0)) &&
              cast<ConstantInt>(M->getOperand(0))->isNullValue() &&
              M->getOperand(1) &&
              isa<ConstantInt>(M->getOperand(1)) &&
              cast<ConstantInt>(M->getOperand(1))->getValue() == Size &&
              M->getOperand(2) &&
              isa<MDNode>(M->getOperand(2)))
            CopyMD = cast<MDNode>(M->getOperand(2));
        }
      }
    }
  }

  // If the memcpy/memmove provides better alignment info than we can
  // infer, use it.
  SrcAlign = std::max(SrcAlign, CopyAlign);
  DstAlign = std::max(DstAlign, CopyAlign);

  Value *Src = Builder->CreateBitCast(MI->getArgOperand(1), NewSrcPtrTy);
  Value *Dest = Builder->CreateBitCast(MI->getArgOperand(0), NewDstPtrTy);
  LoadInst *L = Builder->CreateLoad(Src, MI->isVolatile());
  L->setAlignment(SrcAlign);
  if (CopyMD)
    L->setMetadata(LLVMContext::MD_tbaa, CopyMD);
  StoreInst *S = Builder->CreateStore(L, Dest, MI->isVolatile());
  S->setAlignment(DstAlign);
  if (CopyMD)
    S->setMetadata(LLVMContext::MD_tbaa, CopyMD);

  // Set the size of the copy to 0, it will be deleted on the next iteration.
  MI->setArgOperand(2, Constant::getNullValue(MemOpLength->getType()));
  return MI;
}

Value * InstCombiner::SimplifyShrShlDemandedBits ( Instruction *  Shr, Instruction *  Shl, APInt  DemandedMask, APInt &  KnownZero, APInt &  KnownOne  )

private

Helper routine of SimplifyDemandedUseBits. It tries to simplify demanded bit for "r1 = shr x, c1; r2 = shl r1, c2" instruction sequence.

Helper routine of SimplifyDemandedUseBits. It tries to simplify "E1 = (X lsr C1) << C2", where the C1 and C2 are constant, into "E2 = X << (C2 - C1)" or "E2 = X >> (C1 - C2)", depending on the sign of "C2-C1".

Suppose E1 and E2 are generally different in bits S={bm, bm+1, ..., bn}, without considering the specific value X is holding. This transformation is legal iff one of following conditions is hold: 1) All the bit in S are 0, in this case E1 == E2. 2) We don't care those bits in S, per the input DemandedMask. 3) Combination of 1) and 2). Some bits in S are 0, and we don't care the rest bits.

Currently we only test condition 2).

As with SimplifyDemandedUseBits, it returns NULL if the simplification was not successful.

Definition at line 845 of file InstCombineSimplifyDemanded.cpp.

                                                                           {

  const APInt &ShlOp1 = cast<ConstantInt>(Shl->getOperand(1))->getValue();
  const APInt &ShrOp1 = cast<ConstantInt>(Shr->getOperand(1))->getValue();
  if (!ShlOp1 || !ShrOp1)
      return nullptr; // Noop.

  Value *VarX = Shr->getOperand(0);
  Type *Ty = VarX->getType();
  unsigned BitWidth = Ty->getIntegerBitWidth();
  if (ShlOp1.uge(BitWidth) || ShrOp1.uge(BitWidth))
    return nullptr; // Undef.

  unsigned ShlAmt = ShlOp1.getZExtValue();
  unsigned ShrAmt = ShrOp1.getZExtValue();

  KnownOne.clearAllBits();
  KnownZero = APInt::getBitsSet(KnownZero.getBitWidth(), 0, ShlAmt-1);
  KnownZero &= DemandedMask;

  APInt BitMask1(APInt::getAllOnesValue(BitWidth));
  APInt BitMask2(APInt::getAllOnesValue(BitWidth));

  bool isLshr = (Shr->getOpcode() == Instruction::LShr);
  BitMask1 = isLshr ? (BitMask1.lshr(ShrAmt) << ShlAmt) :
                      (BitMask1.ashr(ShrAmt) << ShlAmt);

  if (ShrAmt <= ShlAmt) {
    BitMask2 <<= (ShlAmt - ShrAmt);
  } else {
    BitMask2 = isLshr ? BitMask2.lshr(ShrAmt - ShlAmt):
                        BitMask2.ashr(ShrAmt - ShlAmt);
  }

  // Check if condition-2 (see the comment to this function) is satified.
  if ((BitMask1 & DemandedMask) == (BitMask2 & DemandedMask)) {
    if (ShrAmt == ShlAmt)
      return VarX;

    if (!Shr->hasOneUse())
      return nullptr;

    BinaryOperator *New;
    if (ShrAmt < ShlAmt) {
      Constant *Amt = ConstantInt::get(VarX->getType(), ShlAmt - ShrAmt);
      New = BinaryOperator::CreateShl(VarX, Amt);
      BinaryOperator *Orig = cast<BinaryOperator>(Shl);
      New->setHasNoSignedWrap(Orig->hasNoSignedWrap());
      New->setHasNoUnsignedWrap(Orig->hasNoUnsignedWrap());
    } else {
      Constant *Amt = ConstantInt::get(VarX->getType(), ShrAmt - ShlAmt);
      New = isLshr ? BinaryOperator::CreateLShr(VarX, Amt) :
                     BinaryOperator::CreateAShr(VarX, Amt);
      if (cast<BinaryOperator>(Shr)->isExact())
        New->setIsExact(true);
    }

    return InsertNewInstWith(New, *Shl);
  }

  return nullptr;
}

bool InstCombiner::SimplifyStoreAtEndOfBlock ( StoreInst &  SI )

private

SimplifyStoreAtEndOfBlock - Turn things like: if () { *P = v1; } else { *P = v2 } into a phi node with a store in the successor.

Simplify things like: *P = v1; if () { *P = v2; } into a phi node with a store in the successor.

Definition at line 726 of file InstCombineLoadStoreAlloca.cpp.

                                                          {
  BasicBlock *StoreBB = SI.getParent();

  // Check to see if the successor block has exactly two incoming edges.  If
  // so, see if the other predecessor contains a store to the same location.
  // if so, insert a PHI node (if needed) and move the stores down.
  BasicBlock *DestBB = StoreBB->getTerminator()->getSuccessor(0);

  // Determine whether Dest has exactly two predecessors and, if so, compute
  // the other predecessor.
  pred_iterator PI = pred_begin(DestBB);
  BasicBlock *P = *PI;
  BasicBlock *OtherBB = nullptr;

  if (P != StoreBB)
    OtherBB = P;

  if (++PI == pred_end(DestBB))
    return false;

  P = *PI;
  if (P != StoreBB) {
    if (OtherBB)
      return false;
    OtherBB = P;
  }
  if (++PI != pred_end(DestBB))
    return false;

  // Bail out if all the relevant blocks aren't distinct (this can happen,
  // for example, if SI is in an infinite loop)
  if (StoreBB == DestBB || OtherBB == DestBB)
    return false;

  // Verify that the other block ends in a branch and is not otherwise empty.
  BasicBlock::iterator BBI = OtherBB->getTerminator();
  BranchInst *OtherBr = dyn_cast<BranchInst>(BBI);
  if (!OtherBr || BBI == OtherBB->begin())
    return false;

  // If the other block ends in an unconditional branch, check for the 'if then
  // else' case.  there is an instruction before the branch.
  StoreInst *OtherStore = nullptr;
  if (OtherBr->isUnconditional()) {
    --BBI;
    // Skip over debugging info.
    while (isa<DbgInfoIntrinsic>(BBI) ||
           (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
      if (BBI==OtherBB->begin())
        return false;
      --BBI;
    }
    // If this isn't a store, isn't a store to the same location, or is not the
    // right kind of store, bail out.
    OtherStore = dyn_cast<StoreInst>(BBI);
    if (!OtherStore || OtherStore->getOperand(1) != SI.getOperand(1) ||
        !SI.isSameOperationAs(OtherStore))
      return false;
  } else {
    // Otherwise, the other block ended with a conditional branch. If one of the
    // destinations is StoreBB, then we have the if/then case.
    if (OtherBr->getSuccessor(0) != StoreBB &&
        OtherBr->getSuccessor(1) != StoreBB)
      return false;

    // Okay, we know that OtherBr now goes to Dest and StoreBB, so this is an
    // if/then triangle.  See if there is a store to the same ptr as SI that
    // lives in OtherBB.
    for (;; --BBI) {
      // Check to see if we find the matching store.
      if ((OtherStore = dyn_cast<StoreInst>(BBI))) {
        if (OtherStore->getOperand(1) != SI.getOperand(1) ||
            !SI.isSameOperationAs(OtherStore))
          return false;
        break;
      }
      // If we find something that may be using or overwriting the stored
      // value, or if we run out of instructions, we can't do the xform.
      if (BBI->mayReadFromMemory() || BBI->mayWriteToMemory() ||
          BBI == OtherBB->begin())
        return false;
    }

    // In order to eliminate the store in OtherBr, we have to
    // make sure nothing reads or overwrites the stored value in
    // StoreBB.
    for (BasicBlock::iterator I = StoreBB->begin(); &*I != &SI; ++I) {
      // FIXME: This should really be AA driven.
      if (I->mayReadFromMemory() || I->mayWriteToMemory())
        return false;
    }
  }

  // Insert a PHI node now if we need it.
  Value *MergedVal = OtherStore->getOperand(0);
  if (MergedVal != SI.getOperand(0)) {
    PHINode *PN = PHINode::Create(MergedVal->getType(), 2, "storemerge");
    PN->addIncoming(SI.getOperand(0), SI.getParent());
    PN->addIncoming(OtherStore->getOperand(0), OtherBB);
    MergedVal = InsertNewInstBefore(PN, DestBB->front());
  }

  // Advance to a place where it is safe to insert the new store and
  // insert it.
  BBI = DestBB->getFirstInsertionPt();
  StoreInst *NewSI = new StoreInst(MergedVal, SI.getOperand(1),
                                   SI.isVolatile(),
                                   SI.getAlignment(),
                                   SI.getOrdering(),
                                   SI.getSynchScope());
  InsertNewInstBefore(NewSI, *BBI);
  NewSI->setDebugLoc(OtherStore->getDebugLoc());

  // If the two stores had the same TBAA tag, preserve it.
  if (MDNode *TBAATag = SI.getMetadata(LLVMContext::MD_tbaa))
    if ((TBAATag = MDNode::getMostGenericTBAA(TBAATag,
                               OtherStore->getMetadata(LLVMContext::MD_tbaa))))
      NewSI->setMetadata(LLVMContext::MD_tbaa, TBAATag);


  // Nuke the old stores.
  EraseInstFromFunction(SI);
  EraseInstFromFunction(*OtherStore);
  return true;
}

Value * InstCombiner::SimplifyUsingDistributiveLaws ( BinaryOperator &  I )

private

SimplifyUsingDistributiveLaws - This tries to simplify binary operations which some other binary operation distributes over either by factorizing out common terms (eg "(A*B)+(A*C)" -> "A*(B+C)") or expanding out if this results in simplifications (eg: "A & (B | C) -> (A&B) | (A&C)" if this is a win). Returns the simplified value, or null if it didn't simplify.

Definition at line 525 of file InstructionCombining.cpp.

                                                                    {
  Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
  BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS);
  BinaryOperator *Op1 = dyn_cast<BinaryOperator>(RHS);

  // Factorization.
  Value *A = nullptr, *B = nullptr, *C = nullptr, *D = nullptr;
  Instruction::BinaryOps LHSOpcode = getBinOpsForFactorization(Op0, A, B);
  Instruction::BinaryOps RHSOpcode = getBinOpsForFactorization(Op1, C, D);

  // The instruction has the form "(A op' B) op (C op' D)".  Try to factorize
  // a common term.
  if (LHSOpcode == RHSOpcode) {
    if (Value *V = tryFactorization(Builder, DL, I, LHSOpcode, A, B, C, D))
      return V;
  }

  // The instruction has the form "(A op' B) op (C)".  Try to factorize common
  // term.
  if (Value *V = tryFactorization(Builder, DL, I, LHSOpcode, A, B, RHS,
                                  getIdentityValue(LHSOpcode, RHS)))
    return V;

  // The instruction has the form "(B) op (C op' D)".  Try to factorize common
  // term.
  if (Value *V = tryFactorization(Builder, DL, I, RHSOpcode, LHS,
                                  getIdentityValue(RHSOpcode, LHS), C, D))
    return V;

  // Expansion.
  Instruction::BinaryOps TopLevelOpcode = I.getOpcode();
  if (Op0 && RightDistributesOverLeft(Op0->getOpcode(), TopLevelOpcode)) {
    // The instruction has the form "(A op' B) op C".  See if expanding it out
    // to "(A op C) op' (B op C)" results in simplifications.
    Value *A = Op0->getOperand(0), *B = Op0->getOperand(1), *C = RHS;
    Instruction::BinaryOps InnerOpcode = Op0->getOpcode(); // op'

    // Do "A op C" and "B op C" both simplify?
    if (Value *L = SimplifyBinOp(TopLevelOpcode, A, C, DL))
      if (Value *R = SimplifyBinOp(TopLevelOpcode, B, C, DL)) {
        // They do! Return "L op' R".
        ++NumExpand;
        // If "L op' R" equals "A op' B" then "L op' R" is just the LHS.
        if ((L == A && R == B) ||
            (Instruction::isCommutative(InnerOpcode) && L == B && R == A))
          return Op0;
        // Otherwise return "L op' R" if it simplifies.
        if (Value *V = SimplifyBinOp(InnerOpcode, L, R, DL))
          return V;
        // Otherwise, create a new instruction.
        C = Builder->CreateBinOp(InnerOpcode, L, R);
        C->takeName(&I);
        return C;
      }
  }

  if (Op1 && LeftDistributesOverRight(TopLevelOpcode, Op1->getOpcode())) {
    // The instruction has the form "A op (B op' C)".  See if expanding it out
    // to "(A op B) op' (A op C)" results in simplifications.
    Value *A = LHS, *B = Op1->getOperand(0), *C = Op1->getOperand(1);
    Instruction::BinaryOps InnerOpcode = Op1->getOpcode(); // op'

    // Do "A op B" and "A op C" both simplify?
    if (Value *L = SimplifyBinOp(TopLevelOpcode, A, B, DL))
      if (Value *R = SimplifyBinOp(TopLevelOpcode, A, C, DL)) {
        // They do! Return "L op' R".
        ++NumExpand;
        // If "L op' R" equals "B op' C" then "L op' R" is just the RHS.
        if ((L == B && R == C) ||
            (Instruction::isCommutative(InnerOpcode) && L == C && R == B))
          return Op1;
        // Otherwise return "L op' R" if it simplifies.
        if (Value *V = SimplifyBinOp(InnerOpcode, L, R, DL))
          return V;
        // Otherwise, create a new instruction.
        A = Builder->CreateBinOp(InnerOpcode, L, R);
        A->takeName(&I);
        return A;
      }
  }

  return nullptr;
}

Value * InstCombiner::SimplifyVectorOp ( BinaryOperator &  Inst )

private

Makes transformation of binary operation specific for vector types.

Parameters

Inst	Binary operator to transform.

Returns

Pointer to node that must replace the original binary operator, or null pointer if no transformation was made.

Definition at line 1202 of file InstructionCombining.cpp.

                                                          {
  if (!Inst.getType()->isVectorTy()) return nullptr;

  // It may not be safe to reorder shuffles and things like div, urem, etc.
  // because we may trap when executing those ops on unknown vector elements.
  // See PR20059.
  if (!isSafeToSpeculativelyExecute(&Inst, DL)) return nullptr;

  unsigned VWidth = cast<VectorType>(Inst.getType())->getNumElements();
  Value *LHS = Inst.getOperand(0), *RHS = Inst.getOperand(1);
  assert(cast<VectorType>(LHS->getType())->getNumElements() == VWidth);
  assert(cast<VectorType>(RHS->getType())->getNumElements() == VWidth);

  // If both arguments of binary operation are shuffles, which use the same
  // mask and shuffle within a single vector, it is worthwhile to move the
  // shuffle after binary operation:
  //   Op(shuffle(v1, m), shuffle(v2, m)) -> shuffle(Op(v1, v2), m)
  if (isa<ShuffleVectorInst>(LHS) && isa<ShuffleVectorInst>(RHS)) {
    ShuffleVectorInst *LShuf = cast<ShuffleVectorInst>(LHS);
    ShuffleVectorInst *RShuf = cast<ShuffleVectorInst>(RHS);
    if (isa<UndefValue>(LShuf->getOperand(1)) &&
        isa<UndefValue>(RShuf->getOperand(1)) &&
        LShuf->getOperand(0)->getType() == RShuf->getOperand(0)->getType() &&
        LShuf->getMask() == RShuf->getMask()) {
      Value *NewBO = CreateBinOpAsGiven(Inst, LShuf->getOperand(0),
          RShuf->getOperand(0), Builder);
      Value *Res = Builder->CreateShuffleVector(NewBO,
          UndefValue::get(NewBO->getType()), LShuf->getMask());
      return Res;
    }
  }

  // If one argument is a shuffle within one vector, the other is a constant,
  // try moving the shuffle after the binary operation.
  ShuffleVectorInst *Shuffle = nullptr;
  Constant *C1 = nullptr;
  if (isa<ShuffleVectorInst>(LHS)) Shuffle = cast<ShuffleVectorInst>(LHS);
  if (isa<ShuffleVectorInst>(RHS)) Shuffle = cast<ShuffleVectorInst>(RHS);
  if (isa<Constant>(LHS)) C1 = cast<Constant>(LHS);
  if (isa<Constant>(RHS)) C1 = cast<Constant>(RHS);
  if (Shuffle && C1 &&
      (isa<ConstantVector>(C1) || isa<ConstantDataVector>(C1)) &&
      isa<UndefValue>(Shuffle->getOperand(1)) &&
      Shuffle->getType() == Shuffle->getOperand(0)->getType()) {
    SmallVector<int, 16> ShMask = Shuffle->getShuffleMask();
    // Find constant C2 that has property:
    //   shuffle(C2, ShMask) = C1
    // If such constant does not exist (example: ShMask=<0,0> and C1=<1,2>)
    // reorder is not possible.
    SmallVector<Constant*, 16> C2M(VWidth,
                               UndefValue::get(C1->getType()->getScalarType()));
    bool MayChange = true;
    for (unsigned I = 0; I < VWidth; ++I) {
      if (ShMask[I] >= 0) {
        assert(ShMask[I] < (int)VWidth);
        if (!isa<UndefValue>(C2M[ShMask[I]])) {
          MayChange = false;
          break;
        }
        C2M[ShMask[I]] = C1->getAggregateElement(I);
      }
    }
    if (MayChange) {
      Constant *C2 = ConstantVector::get(C2M);
      Value *NewLHS, *NewRHS;
      if (isa<Constant>(LHS)) {
        NewLHS = C2;
        NewRHS = Shuffle->getOperand(0);
      } else {
        NewLHS = Shuffle->getOperand(0);
        NewRHS = C2;
      }
      Value *NewBO = CreateBinOpAsGiven(Inst, NewLHS, NewRHS, Builder);
      Value *Res = Builder->CreateShuffleVector(NewBO,
          UndefValue::get(Inst.getType()), Shuffle->getMask());
      return Res;
    }
  }

  return nullptr;
}

bool FunctionPass::skipOptnoneFunction ( const Function &  F ) const

protectedinherited

skipOptnoneFunction - This function has Attribute::OptimizeNone and most transformation passes should skip it.

Definition at line 143 of file Pass.cpp.

                                                              {
  if (F.hasFnAttribute(Attribute::OptimizeNone)) {
    DEBUG(dbgs() << "Skipping pass '" << getPassName()
          << "' on function " << F.getName() << "\n");
    return true;
  }
  return false;
}

Instruction * InstCombiner::SliceUpIllegalIntegerPHI

(

PHINode &

FirstPhi

)

SliceUpIllegalIntegerPHI - This is an integer PHI and we know that it has an illegal type: see if it is only used by trunc or trunc(lshr) operations. If so, we split the PHI into the various pieces being extracted. This sort of thing is introduced when SROA promotes an aggregate to large integer values.

TODO: The user of the trunc may be an bitcast to float/double/vector or an inttoptr. We should produce new PHIs in the right type.

Definition at line 618 of file InstCombinePHI.cpp.

                                                                     {
  // PHIUsers - Keep track of all of the truncated values extracted from a set
  // of PHIs, along with their offset.  These are the things we want to rewrite.
  SmallVector<PHIUsageRecord, 16> PHIUsers;

  // PHIs are often mutually cyclic, so we keep track of a whole set of PHI
  // nodes which are extracted from. PHIsToSlice is a set we use to avoid
  // revisiting PHIs, PHIsInspected is a ordered list of PHIs that we need to
  // check the uses of (to ensure they are all extracts).
  SmallVector<PHINode*, 8> PHIsToSlice;
  SmallPtrSet<PHINode*, 8> PHIsInspected;

  PHIsToSlice.push_back(&FirstPhi);
  PHIsInspected.insert(&FirstPhi);

  for (unsigned PHIId = 0; PHIId != PHIsToSlice.size(); ++PHIId) {
    PHINode *PN = PHIsToSlice[PHIId];

    // Scan the input list of the PHI.  If any input is an invoke, and if the
    // input is defined in the predecessor, then we won't be split the critical
    // edge which is required to insert a truncate.  Because of this, we have to
    // bail out.
    for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
      InvokeInst *II = dyn_cast<InvokeInst>(PN->getIncomingValue(i));
      if (!II) continue;
      if (II->getParent() != PN->getIncomingBlock(i))
        continue;

      // If we have a phi, and if it's directly in the predecessor, then we have
      // a critical edge where we need to put the truncate.  Since we can't
      // split the edge in instcombine, we have to bail out.
      return nullptr;
    }

    for (User *U : PN->users()) {
      Instruction *UserI = cast<Instruction>(U);

      // If the user is a PHI, inspect its uses recursively.
      if (PHINode *UserPN = dyn_cast<PHINode>(UserI)) {
        if (PHIsInspected.insert(UserPN))
          PHIsToSlice.push_back(UserPN);
        continue;
      }

      // Truncates are always ok.
      if (isa<TruncInst>(UserI)) {
        PHIUsers.push_back(PHIUsageRecord(PHIId, 0, UserI));
        continue;
      }

      // Otherwise it must be a lshr which can only be used by one trunc.
      if (UserI->getOpcode() != Instruction::LShr ||
          !UserI->hasOneUse() || !isa<TruncInst>(UserI->user_back()) ||
          !isa<ConstantInt>(UserI->getOperand(1)))
        return nullptr;

      unsigned Shift = cast<ConstantInt>(UserI->getOperand(1))->getZExtValue();
      PHIUsers.push_back(PHIUsageRecord(PHIId, Shift, UserI->user_back()));
    }
  }

  // If we have no users, they must be all self uses, just nuke the PHI.
  if (PHIUsers.empty())
    return ReplaceInstUsesWith(FirstPhi, UndefValue::get(FirstPhi.getType()));

  // If this phi node is transformable, create new PHIs for all the pieces
  // extracted out of it.  First, sort the users by their offset and size.
  array_pod_sort(PHIUsers.begin(), PHIUsers.end());

  DEBUG(dbgs() << "SLICING UP PHI: " << FirstPhi << '\n';
        for (unsigned i = 1, e = PHIsToSlice.size(); i != e; ++i)
          dbgs() << "AND USER PHI #" << i << ": " << *PHIsToSlice[i] << '\n';
    );

  // PredValues - This is a temporary used when rewriting PHI nodes.  It is
  // hoisted out here to avoid construction/destruction thrashing.
  DenseMap<BasicBlock*, Value*> PredValues;

  // ExtractedVals - Each new PHI we introduce is saved here so we don't
  // introduce redundant PHIs.
  DenseMap<LoweredPHIRecord, PHINode*> ExtractedVals;

  for (unsigned UserI = 0, UserE = PHIUsers.size(); UserI != UserE; ++UserI) {
    unsigned PHIId = PHIUsers[UserI].PHIId;
    PHINode *PN = PHIsToSlice[PHIId];
    unsigned Offset = PHIUsers[UserI].Shift;
    Type *Ty = PHIUsers[UserI].Inst->getType();

    PHINode *EltPHI;

    // If we've already lowered a user like this, reuse the previously lowered
    // value.
    if ((EltPHI = ExtractedVals[LoweredPHIRecord(PN, Offset, Ty)]) == nullptr) {

      // Otherwise, Create the new PHI node for this user.
      EltPHI = PHINode::Create(Ty, PN->getNumIncomingValues(),
                               PN->getName()+".off"+Twine(Offset), PN);
      assert(EltPHI->getType() != PN->getType() &&
             "Truncate didn't shrink phi?");

      for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
        BasicBlock *Pred = PN->getIncomingBlock(i);
        Value *&PredVal = PredValues[Pred];

        // If we already have a value for this predecessor, reuse it.
        if (PredVal) {
          EltPHI->addIncoming(PredVal, Pred);
          continue;
        }

        // Handle the PHI self-reuse case.
        Value *InVal = PN->getIncomingValue(i);
        if (InVal == PN) {
          PredVal = EltPHI;
          EltPHI->addIncoming(PredVal, Pred);
          continue;
        }

        if (PHINode *InPHI = dyn_cast<PHINode>(PN)) {
          // If the incoming value was a PHI, and if it was one of the PHIs we
          // already rewrote it, just use the lowered value.
          if (Value *Res = ExtractedVals[LoweredPHIRecord(InPHI, Offset, Ty)]) {
            PredVal = Res;
            EltPHI->addIncoming(PredVal, Pred);
            continue;
          }
        }

        // Otherwise, do an extract in the predecessor.
        Builder->SetInsertPoint(Pred, Pred->getTerminator());
        Value *Res = InVal;
        if (Offset)
          Res = Builder->CreateLShr(Res, ConstantInt::get(InVal->getType(),
                                                          Offset), "extract");
        Res = Builder->CreateTrunc(Res, Ty, "extract.t");
        PredVal = Res;
        EltPHI->addIncoming(Res, Pred);

        // If the incoming value was a PHI, and if it was one of the PHIs we are
        // rewriting, we will ultimately delete the code we inserted.  This
        // means we need to revisit that PHI to make sure we extract out the
        // needed piece.
        if (PHINode *OldInVal = dyn_cast<PHINode>(PN->getIncomingValue(i)))
          if (PHIsInspected.count(OldInVal)) {
            unsigned RefPHIId = std::find(PHIsToSlice.begin(),PHIsToSlice.end(),
                                          OldInVal)-PHIsToSlice.begin();
            PHIUsers.push_back(PHIUsageRecord(RefPHIId, Offset,
                                              cast<Instruction>(Res)));
            ++UserE;
          }
      }
      PredValues.clear();

      DEBUG(dbgs() << "  Made element PHI for offset " << Offset << ": "
                   << *EltPHI << '\n');
      ExtractedVals[LoweredPHIRecord(PN, Offset, Ty)] = EltPHI;
    }

    // Replace the use of this piece with the PHI node.
    ReplaceInstUsesWith(*PHIUsers[UserI].Inst, EltPHI);
  }

  // Replace all the remaining uses of the PHI nodes (self uses and the lshrs)
  // with undefs.
  Value *Undef = UndefValue::get(FirstPhi.getType());
  for (unsigned i = 1, e = PHIsToSlice.size(); i != e; ++i)
    ReplaceInstUsesWith(*PHIsToSlice[i], Undef);
  return ReplaceInstUsesWith(FirstPhi, Undef);
}

Instruction * InstCombiner::transformCallThroughTrampoline ( CallSite  CS, IntrinsicInst *  Tramp  )

private

Definition at line 1487 of file InstCombineCalls.cpp.

                                                                   {
  Value *Callee = CS.getCalledValue();
  PointerType *PTy = cast<PointerType>(Callee->getType());
  FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
  const AttributeSet &Attrs = CS.getAttributes();

  // If the call already has the 'nest' attribute somewhere then give up -
  // otherwise 'nest' would occur twice after splicing in the chain.
  if (Attrs.hasAttrSomewhere(Attribute::Nest))
    return nullptr;

  assert(Tramp &&
         "transformCallThroughTrampoline called with incorrect CallSite.");

  Function *NestF =cast<Function>(Tramp->getArgOperand(1)->stripPointerCasts());
  PointerType *NestFPTy = cast<PointerType>(NestF->getType());
  FunctionType *NestFTy = cast<FunctionType>(NestFPTy->getElementType());

  const AttributeSet &NestAttrs = NestF->getAttributes();
  if (!NestAttrs.isEmpty()) {
    unsigned NestIdx = 1;
    Type *NestTy = nullptr;
    AttributeSet NestAttr;

    // Look for a parameter marked with the 'nest' attribute.
    for (FunctionType::param_iterator I = NestFTy->param_begin(),
         E = NestFTy->param_end(); I != E; ++NestIdx, ++I)
      if (NestAttrs.hasAttribute(NestIdx, Attribute::Nest)) {
        // Record the parameter type and any other attributes.
        NestTy = *I;
        NestAttr = NestAttrs.getParamAttributes(NestIdx);
        break;
      }

    if (NestTy) {
      Instruction *Caller = CS.getInstruction();
      std::vector<Value*> NewArgs;
      NewArgs.reserve(CS.arg_size() + 1);

      SmallVector<AttributeSet, 8> NewAttrs;
      NewAttrs.reserve(Attrs.getNumSlots() + 1);

      // Insert the nest argument into the call argument list, which may
      // mean appending it.  Likewise for attributes.

      // Add any result attributes.
      if (Attrs.hasAttributes(AttributeSet::ReturnIndex))
        NewAttrs.push_back(AttributeSet::get(Caller->getContext(),
                                             Attrs.getRetAttributes()));

      {
        unsigned Idx = 1;
        CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end();
        do {
          if (Idx == NestIdx) {
            // Add the chain argument and attributes.
            Value *NestVal = Tramp->getArgOperand(2);
            if (NestVal->getType() != NestTy)
              NestVal = Builder->CreateBitCast(NestVal, NestTy, "nest");
            NewArgs.push_back(NestVal);
            NewAttrs.push_back(AttributeSet::get(Caller->getContext(),
                                                 NestAttr));
          }

          if (I == E)
            break;

          // Add the original argument and attributes.
          NewArgs.push_back(*I);
          AttributeSet Attr = Attrs.getParamAttributes(Idx);
          if (Attr.hasAttributes(Idx)) {
            AttrBuilder B(Attr, Idx);
            NewAttrs.push_back(AttributeSet::get(Caller->getContext(),
                                                 Idx + (Idx >= NestIdx), B));
          }

          ++Idx, ++I;
        } while (1);
      }

      // Add any function attributes.
      if (Attrs.hasAttributes(AttributeSet::FunctionIndex))
        NewAttrs.push_back(AttributeSet::get(FTy->getContext(),
                                             Attrs.getFnAttributes()));

      // The trampoline may have been bitcast to a bogus type (FTy).
      // Handle this by synthesizing a new function type, equal to FTy
      // with the chain parameter inserted.

      std::vector<Type*> NewTypes;
      NewTypes.reserve(FTy->getNumParams()+1);

      // Insert the chain's type into the list of parameter types, which may
      // mean appending it.
      {
        unsigned Idx = 1;
        FunctionType::param_iterator I = FTy->param_begin(),
          E = FTy->param_end();

        do {
          if (Idx == NestIdx)
            // Add the chain's type.
            NewTypes.push_back(NestTy);

          if (I == E)
            break;

          // Add the original type.
          NewTypes.push_back(*I);

          ++Idx, ++I;
        } while (1);
      }

      // Replace the trampoline call with a direct call.  Let the generic
      // code sort out any function type mismatches.
      FunctionType *NewFTy = FunctionType::get(FTy->getReturnType(), NewTypes,
                                                FTy->isVarArg());
      Constant *NewCallee =
        NestF->getType() == PointerType::getUnqual(NewFTy) ?
        NestF : ConstantExpr::getBitCast(NestF,
                                         PointerType::getUnqual(NewFTy));
      const AttributeSet &NewPAL =
          AttributeSet::get(FTy->getContext(), NewAttrs);

      Instruction *NewCaller;
      if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
        NewCaller = InvokeInst::Create(NewCallee,
                                       II->getNormalDest(), II->getUnwindDest(),
                                       NewArgs);
        cast<InvokeInst>(NewCaller)->setCallingConv(II->getCallingConv());
        cast<InvokeInst>(NewCaller)->setAttributes(NewPAL);
      } else {
        NewCaller = CallInst::Create(NewCallee, NewArgs);
        if (cast<CallInst>(Caller)->isTailCall())
          cast<CallInst>(NewCaller)->setTailCall();
        cast<CallInst>(NewCaller)->
          setCallingConv(cast<CallInst>(Caller)->getCallingConv());
        cast<CallInst>(NewCaller)->setAttributes(NewPAL);
      }

      return NewCaller;
    }
  }

  // Replace the trampoline call with a direct call.  Since there is no 'nest'
  // parameter, there is no need to adjust the argument list.  Let the generic
  // code sort out any function type mismatches.
  Constant *NewCallee =
    NestF->getType() == PTy ? NestF :
                              ConstantExpr::getBitCast(NestF, PTy);
  CS.setCalledFunction(NewCallee);
  return CS.getInstruction();
}

bool InstCombiner::transformConstExprCastCall ( CallSite  CS )

private

Definition at line 1227 of file InstCombineCalls.cpp.

                                                         {
  Function *Callee =
    dyn_cast<Function>(CS.getCalledValue()->stripPointerCasts());
  if (!Callee)
    return false;
  Instruction *Caller = CS.getInstruction();
  const AttributeSet &CallerPAL = CS.getAttributes();

  // Okay, this is a cast from a function to a different type.  Unless doing so
  // would cause a type conversion of one of our arguments, change this call to
  // be a direct call with arguments casted to the appropriate types.
  //
  FunctionType *FT = Callee->getFunctionType();
  Type *OldRetTy = Caller->getType();
  Type *NewRetTy = FT->getReturnType();

  // Check to see if we are changing the return type...
  if (OldRetTy != NewRetTy) {

    if (NewRetTy->isStructTy())
      return false; // TODO: Handle multiple return values.

    if (!CastInst::isBitCastable(NewRetTy, OldRetTy)) {
      if (Callee->isDeclaration())
        return false;   // Cannot transform this return value.

      if (!Caller->use_empty() &&
          // void -> non-void is handled specially
          !NewRetTy->isVoidTy())
      return false;   // Cannot transform this return value.
    }

    if (!CallerPAL.isEmpty() && !Caller->use_empty()) {
      AttrBuilder RAttrs(CallerPAL, AttributeSet::ReturnIndex);
      if (RAttrs.
          hasAttributes(AttributeFuncs::
                        typeIncompatible(NewRetTy, AttributeSet::ReturnIndex),
                        AttributeSet::ReturnIndex))
        return false;   // Attribute not compatible with transformed value.
    }

    // If the callsite is an invoke instruction, and the return value is used by
    // a PHI node in a successor, we cannot change the return type of the call
    // because there is no place to put the cast instruction (without breaking
    // the critical edge).  Bail out in this case.
    if (!Caller->use_empty())
      if (InvokeInst *II = dyn_cast<InvokeInst>(Caller))
        for (User *U : II->users())
          if (PHINode *PN = dyn_cast<PHINode>(U))
            if (PN->getParent() == II->getNormalDest() ||
                PN->getParent() == II->getUnwindDest())
              return false;
  }

  unsigned NumActualArgs = CS.arg_size();
  unsigned NumCommonArgs = std::min(FT->getNumParams(), NumActualArgs);

  CallSite::arg_iterator AI = CS.arg_begin();
  for (unsigned i = 0, e = NumCommonArgs; i != e; ++i, ++AI) {
    Type *ParamTy = FT->getParamType(i);
    Type *ActTy = (*AI)->getType();

    if (!CastInst::isBitCastable(ActTy, ParamTy))
      return false;   // Cannot transform this parameter value.

    if (AttrBuilder(CallerPAL.getParamAttributes(i + 1), i + 1).
          hasAttributes(AttributeFuncs::
                        typeIncompatible(ParamTy, i + 1), i + 1))
      return false;   // Attribute not compatible with transformed value.

    if (CS.isInAllocaArgument(i))
      return false;   // Cannot transform to and from inalloca.

    // If the parameter is passed as a byval argument, then we have to have a
    // sized type and the sized type has to have the same size as the old type.
    if (ParamTy != ActTy &&
        CallerPAL.getParamAttributes(i + 1).hasAttribute(i + 1,
                                                         Attribute::ByVal)) {
      PointerType *ParamPTy = dyn_cast<PointerType>(ParamTy);
      if (!ParamPTy || !ParamPTy->getElementType()->isSized() || !DL)
        return false;

      Type *CurElTy = ActTy->getPointerElementType();
      if (DL->getTypeAllocSize(CurElTy) !=
          DL->getTypeAllocSize(ParamPTy->getElementType()))
        return false;
    }
  }

  if (Callee->isDeclaration()) {
    // Do not delete arguments unless we have a function body.
    if (FT->getNumParams() < NumActualArgs && !FT->isVarArg())
      return false;

    // If the callee is just a declaration, don't change the varargsness of the
    // call.  We don't want to introduce a varargs call where one doesn't
    // already exist.
    PointerType *APTy = cast<PointerType>(CS.getCalledValue()->getType());
    if (FT->isVarArg()!=cast<FunctionType>(APTy->getElementType())->isVarArg())
      return false;

    // If both the callee and the cast type are varargs, we still have to make
    // sure the number of fixed parameters are the same or we have the same
    // ABI issues as if we introduce a varargs call.
    if (FT->isVarArg() &&
        cast<FunctionType>(APTy->getElementType())->isVarArg() &&
        FT->getNumParams() !=
        cast<FunctionType>(APTy->getElementType())->getNumParams())
      return false;
  }

  if (FT->getNumParams() < NumActualArgs && FT->isVarArg() &&
      !CallerPAL.isEmpty())
    // In this case we have more arguments than the new function type, but we
    // won't be dropping them.  Check that these extra arguments have attributes
    // that are compatible with being a vararg call argument.
    for (unsigned i = CallerPAL.getNumSlots(); i; --i) {
      unsigned Index = CallerPAL.getSlotIndex(i - 1);
      if (Index <= FT->getNumParams())
        break;

      // Check if it has an attribute that's incompatible with varargs.
      AttributeSet PAttrs = CallerPAL.getSlotAttributes(i - 1);
      if (PAttrs.hasAttribute(Index, Attribute::StructRet))
        return false;
    }


  // Okay, we decided that this is a safe thing to do: go ahead and start
  // inserting cast instructions as necessary.
  std::vector<Value*> Args;
  Args.reserve(NumActualArgs);
  SmallVector<AttributeSet, 8> attrVec;
  attrVec.reserve(NumCommonArgs);

  // Get any return attributes.
  AttrBuilder RAttrs(CallerPAL, AttributeSet::ReturnIndex);

  // If the return value is not being used, the type may not be compatible
  // with the existing attributes.  Wipe out any problematic attributes.
  RAttrs.
    removeAttributes(AttributeFuncs::
                     typeIncompatible(NewRetTy, AttributeSet::ReturnIndex),
                     AttributeSet::ReturnIndex);

  // Add the new return attributes.
  if (RAttrs.hasAttributes())
    attrVec.push_back(AttributeSet::get(Caller->getContext(),
                                        AttributeSet::ReturnIndex, RAttrs));

  AI = CS.arg_begin();
  for (unsigned i = 0; i != NumCommonArgs; ++i, ++AI) {
    Type *ParamTy = FT->getParamType(i);

    if ((*AI)->getType() == ParamTy) {
      Args.push_back(*AI);
    } else {
      Args.push_back(Builder->CreateBitCast(*AI, ParamTy));
    }

    // Add any parameter attributes.
    AttrBuilder PAttrs(CallerPAL.getParamAttributes(i + 1), i + 1);
    if (PAttrs.hasAttributes())
      attrVec.push_back(AttributeSet::get(Caller->getContext(), i + 1,
                                          PAttrs));
  }

  // If the function takes more arguments than the call was taking, add them
  // now.
  for (unsigned i = NumCommonArgs; i != FT->getNumParams(); ++i)
    Args.push_back(Constant::getNullValue(FT->getParamType(i)));

  // If we are removing arguments to the function, emit an obnoxious warning.
  if (FT->getNumParams() < NumActualArgs) {
    // TODO: if (!FT->isVarArg()) this call may be unreachable. PR14722
    if (FT->isVarArg()) {
      // Add all of the arguments in their promoted form to the arg list.
      for (unsigned i = FT->getNumParams(); i != NumActualArgs; ++i, ++AI) {
        Type *PTy = getPromotedType((*AI)->getType());
        if (PTy != (*AI)->getType()) {
          // Must promote to pass through va_arg area!
          Instruction::CastOps opcode =
            CastInst::getCastOpcode(*AI, false, PTy, false);
          Args.push_back(Builder->CreateCast(opcode, *AI, PTy));
        } else {
          Args.push_back(*AI);
        }

        // Add any parameter attributes.
        AttrBuilder PAttrs(CallerPAL.getParamAttributes(i + 1), i + 1);
        if (PAttrs.hasAttributes())
          attrVec.push_back(AttributeSet::get(FT->getContext(), i + 1,
                                              PAttrs));
      }
    }
  }

  AttributeSet FnAttrs = CallerPAL.getFnAttributes();
  if (CallerPAL.hasAttributes(AttributeSet::FunctionIndex))
    attrVec.push_back(AttributeSet::get(Callee->getContext(), FnAttrs));

  if (NewRetTy->isVoidTy())
    Caller->setName("");   // Void type should not have a name.

  const AttributeSet &NewCallerPAL = AttributeSet::get(Callee->getContext(),
                                                       attrVec);

  Instruction *NC;
  if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
    NC = Builder->CreateInvoke(Callee, II->getNormalDest(),
                               II->getUnwindDest(), Args);
    NC->takeName(II);
    cast<InvokeInst>(NC)->setCallingConv(II->getCallingConv());
    cast<InvokeInst>(NC)->setAttributes(NewCallerPAL);
  } else {
    CallInst *CI = cast<CallInst>(Caller);
    NC = Builder->CreateCall(Callee, Args);
    NC->takeName(CI);
    if (CI->isTailCall())
      cast<CallInst>(NC)->setTailCall();
    cast<CallInst>(NC)->setCallingConv(CI->getCallingConv());
    cast<CallInst>(NC)->setAttributes(NewCallerPAL);
  }

  // Insert a cast of the return type as necessary.
  Value *NV = NC;
  if (OldRetTy != NV->getType() && !Caller->use_empty()) {
    if (!NV->getType()->isVoidTy()) {
      NV = NC = CastInst::Create(CastInst::BitCast, NC, OldRetTy);
      NC->setDebugLoc(Caller->getDebugLoc());

      // If this is an invoke instruction, we should insert it after the first
      // non-phi, instruction in the normal successor block.
      if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
        BasicBlock::iterator I = II->getNormalDest()->getFirstInsertionPt();
        InsertNewInstBefore(NC, *I);
      } else {
        // Otherwise, it's a call, just insert cast right after the call.
        InsertNewInstBefore(NC, *Caller);
      }
      Worklist.AddUsersToWorkList(*Caller);
    } else {
      NV = UndefValue::get(Caller->getType());
    }
  }

  if (!Caller->use_empty())
    ReplaceInstUsesWith(*Caller, NV);
  else if (Caller->hasValueHandle())
    ValueHandleBase::ValueIsRAUWd(Caller, NV);

  EraseInstFromFunction(*Caller);
  return true;
}

Instruction * InstCombiner::transformSExtICmp ( ICmpInst *  ICI, Instruction &  CI  )

private

transformSExtICmp - Transform (sext icmp) to bitwise / integer operations in order to eliminate the icmp.

Definition at line 894 of file InstCombineCasts.cpp.

                                                                           {
  Value *Op0 = ICI->getOperand(0), *Op1 = ICI->getOperand(1);
  ICmpInst::Predicate Pred = ICI->getPredicate();

  if (Constant *Op1C = dyn_cast<Constant>(Op1)) {
    // (x <s  0) ? -1 : 0 -> ashr x, 31        -> all ones if negative
    // (x >s -1) ? -1 : 0 -> not (ashr x, 31)  -> all ones if positive
    if ((Pred == ICmpInst::ICMP_SLT && Op1C->isNullValue()) ||
        (Pred == ICmpInst::ICMP_SGT && Op1C->isAllOnesValue())) {

      Value *Sh = ConstantInt::get(Op0->getType(),
                                   Op0->getType()->getScalarSizeInBits()-1);
      Value *In = Builder->CreateAShr(Op0, Sh, Op0->getName()+".lobit");
      if (In->getType() != CI.getType())
        In = Builder->CreateIntCast(In, CI.getType(), true/*SExt*/);

      if (Pred == ICmpInst::ICMP_SGT)
        In = Builder->CreateNot(In, In->getName()+".not");
      return ReplaceInstUsesWith(CI, In);
    }
  }

  if (ConstantInt *Op1C = dyn_cast<ConstantInt>(Op1)) {
    // If we know that only one bit of the LHS of the icmp can be set and we
    // have an equality comparison with zero or a power of 2, we can transform
    // the icmp and sext into bitwise/integer operations.
    if (ICI->hasOneUse() &&
        ICI->isEquality() && (Op1C->isZero() || Op1C->getValue().isPowerOf2())){
      unsigned BitWidth = Op1C->getType()->getBitWidth();
      APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
      computeKnownBits(Op0, KnownZero, KnownOne);

      APInt KnownZeroMask(~KnownZero);
      if (KnownZeroMask.isPowerOf2()) {
        Value *In = ICI->getOperand(0);

        // If the icmp tests for a known zero bit we can constant fold it.
        if (!Op1C->isZero() && Op1C->getValue() != KnownZeroMask) {
          Value *V = Pred == ICmpInst::ICMP_NE ?
                       ConstantInt::getAllOnesValue(CI.getType()) :
                       ConstantInt::getNullValue(CI.getType());
          return ReplaceInstUsesWith(CI, V);
        }

        if (!Op1C->isZero() == (Pred == ICmpInst::ICMP_NE)) {
          // sext ((x & 2^n) == 0)   -> (x >> n) - 1
          // sext ((x & 2^n) != 2^n) -> (x >> n) - 1
          unsigned ShiftAmt = KnownZeroMask.countTrailingZeros();
          // Perform a right shift to place the desired bit in the LSB.
          if (ShiftAmt)
            In = Builder->CreateLShr(In,
                                     ConstantInt::get(In->getType(), ShiftAmt));

          // At this point "In" is either 1 or 0. Subtract 1 to turn
          // {1, 0} -> {0, -1}.
          In = Builder->CreateAdd(In,
                                  ConstantInt::getAllOnesValue(In->getType()),
                                  "sext");
        } else {
          // sext ((x & 2^n) != 0)   -> (x << bitwidth-n) a>> bitwidth-1
          // sext ((x & 2^n) == 2^n) -> (x << bitwidth-n) a>> bitwidth-1
          unsigned ShiftAmt = KnownZeroMask.countLeadingZeros();
          // Perform a left shift to place the desired bit in the MSB.
          if (ShiftAmt)
            In = Builder->CreateShl(In,
                                    ConstantInt::get(In->getType(), ShiftAmt));

          // Distribute the bit over the whole bit width.
          In = Builder->CreateAShr(In, ConstantInt::get(In->getType(),
                                                        BitWidth - 1), "sext");
        }

        if (CI.getType() == In->getType())
          return ReplaceInstUsesWith(CI, In);
        return CastInst::CreateIntegerCast(In, CI.getType(), true/*SExt*/);
      }
    }
  }

  return nullptr;
}

Instruction * InstCombiner::transformZExtICmp ( ICmpInst *  ICI, Instruction &  CI, bool  DoXform = true  )

private

transformZExtICmp - Transform (zext icmp) to bitwise / integer operations in order to eliminate the icmp.

Definition at line 513 of file InstCombineCasts.cpp.

                                                           {
  // If we are just checking for a icmp eq of a single bit and zext'ing it
  // to an integer, then shift the bit to the appropriate place and then
  // cast to integer to avoid the comparison.
  if (ConstantInt *Op1C = dyn_cast<ConstantInt>(ICI->getOperand(1))) {
    const APInt &Op1CV = Op1C->getValue();

    // zext (x <s  0) to i32 --> x>>u31      true if signbit set.
    // zext (x >s -1) to i32 --> (x>>u31)^1  true if signbit clear.
    if ((ICI->getPredicate() == ICmpInst::ICMP_SLT && Op1CV == 0) ||
        (ICI->getPredicate() == ICmpInst::ICMP_SGT &&Op1CV.isAllOnesValue())) {
      if (!DoXform) return ICI;

      Value *In = ICI->getOperand(0);
      Value *Sh = ConstantInt::get(In->getType(),
                                   In->getType()->getScalarSizeInBits()-1);
      In = Builder->CreateLShr(In, Sh, In->getName()+".lobit");
      if (In->getType() != CI.getType())
        In = Builder->CreateIntCast(In, CI.getType(), false/*ZExt*/);

      if (ICI->getPredicate() == ICmpInst::ICMP_SGT) {
        Constant *One = ConstantInt::get(In->getType(), 1);
        In = Builder->CreateXor(In, One, In->getName()+".not");
      }

      return ReplaceInstUsesWith(CI, In);
    }

    // zext (X == 0) to i32 --> X^1      iff X has only the low bit set.
    // zext (X == 0) to i32 --> (X>>1)^1 iff X has only the 2nd bit set.
    // zext (X == 1) to i32 --> X        iff X has only the low bit set.
    // zext (X == 2) to i32 --> X>>1     iff X has only the 2nd bit set.
    // zext (X != 0) to i32 --> X        iff X has only the low bit set.
    // zext (X != 0) to i32 --> X>>1     iff X has only the 2nd bit set.
    // zext (X != 1) to i32 --> X^1      iff X has only the low bit set.
    // zext (X != 2) to i32 --> (X>>1)^1 iff X has only the 2nd bit set.
    if ((Op1CV == 0 || Op1CV.isPowerOf2()) &&
        // This only works for EQ and NE
        ICI->isEquality()) {
      // If Op1C some other power of two, convert:
      uint32_t BitWidth = Op1C->getType()->getBitWidth();
      APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
      computeKnownBits(ICI->getOperand(0), KnownZero, KnownOne);

      APInt KnownZeroMask(~KnownZero);
      if (KnownZeroMask.isPowerOf2()) { // Exactly 1 possible 1?
        if (!DoXform) return ICI;

        bool isNE = ICI->getPredicate() == ICmpInst::ICMP_NE;
        if (Op1CV != 0 && (Op1CV != KnownZeroMask)) {
          // (X&4) == 2 --> false
          // (X&4) != 2 --> true
          Constant *Res = ConstantInt::get(Type::getInt1Ty(CI.getContext()),
                                           isNE);
          Res = ConstantExpr::getZExt(Res, CI.getType());
          return ReplaceInstUsesWith(CI, Res);
        }

        uint32_t ShiftAmt = KnownZeroMask.logBase2();
        Value *In = ICI->getOperand(0);
        if (ShiftAmt) {
          // Perform a logical shr by shiftamt.
          // Insert the shift to put the result in the low bit.
          In = Builder->CreateLShr(In, ConstantInt::get(In->getType(),ShiftAmt),
                                   In->getName()+".lobit");
        }

        if ((Op1CV != 0) == isNE) { // Toggle the low bit.
          Constant *One = ConstantInt::get(In->getType(), 1);
          In = Builder->CreateXor(In, One);
        }

        if (CI.getType() == In->getType())
          return ReplaceInstUsesWith(CI, In);
        return CastInst::CreateIntegerCast(In, CI.getType(), false/*ZExt*/);
      }
    }
  }

  // icmp ne A, B is equal to xor A, B when A and B only really have one bit.
  // It is also profitable to transform icmp eq into not(xor(A, B)) because that
  // may lead to additional simplifications.
  if (ICI->isEquality() && CI.getType() == ICI->getOperand(0)->getType()) {
    if (IntegerType *ITy = dyn_cast<IntegerType>(CI.getType())) {
      uint32_t BitWidth = ITy->getBitWidth();
      Value *LHS = ICI->getOperand(0);
      Value *RHS = ICI->getOperand(1);

      APInt KnownZeroLHS(BitWidth, 0), KnownOneLHS(BitWidth, 0);
      APInt KnownZeroRHS(BitWidth, 0), KnownOneRHS(BitWidth, 0);
      computeKnownBits(LHS, KnownZeroLHS, KnownOneLHS);
      computeKnownBits(RHS, KnownZeroRHS, KnownOneRHS);

      if (KnownZeroLHS == KnownZeroRHS && KnownOneLHS == KnownOneRHS) {
        APInt KnownBits = KnownZeroLHS | KnownOneLHS;
        APInt UnknownBit = ~KnownBits;
        if (UnknownBit.countPopulation() == 1) {
          if (!DoXform) return ICI;

          Value *Result = Builder->CreateXor(LHS, RHS);

          // Mask off any bits that are set and won't be shifted away.
          if (KnownOneLHS.uge(UnknownBit))
            Result = Builder->CreateAnd(Result,
                                        ConstantInt::get(ITy, UnknownBit));

          // Shift the bit we're testing down to the lsb.
          Result = Builder->CreateLShr(
               Result, ConstantInt::get(ITy, UnknownBit.countTrailingZeros()));

          if (ICI->getPredicate() == ICmpInst::ICMP_EQ)
            Result = Builder->CreateXor(Result, ConstantInt::get(ITy, 1));
          Result->takeName(ICI);
          return ReplaceInstUsesWith(CI, Result);
        }
      }
    }
  }

  return nullptr;
}

Instruction * InstCombiner::tryOptimizeCall ( CallInst *  CI, const DataLayout *  DL  )

private

Definition at line 1037 of file InstCombineCalls.cpp.

                                                                             {
  if (!CI->getCalledFunction()) return nullptr;

  if (Value *With = Simplifier->optimizeCall(CI)) {
    ++NumSimplified;
    return CI->use_empty() ? CI : ReplaceInstUsesWith(*CI, With);
  }

  return nullptr;
}

void Pass::verifyAnalysis ( ) const

virtualinherited

verifyAnalysis() - This member can be implemented by a analysis pass to check state of analysis information.

Reimplemented in llvm::ScalarEvolution, llvm::RegionInfoPass, llvm::LoopInfo, llvm::DominatorTreeWrapperPass, llvm::MachineRegionInfoPass, and llvm::MachineTraceMetrics.

Definition at line 86 of file Pass.cpp.

                                {
  // By default, don't do anything.
}

void llvm::InstVisitor< InstCombiner , Instruction * >::visit ( Iterator  Start, Iterator  End  )

inlineinherited

Definition at line 90 of file InstVisitor.h.

                                           {
    while (Start != End)
      static_cast<SubClass*>(this)->visit(*Start++);
  }

void llvm::InstVisitor< InstCombiner , Instruction * >::visit ( Module &  M )

inlineinherited

Definition at line 97 of file InstVisitor.h.

                        {
    static_cast<SubClass*>(this)->visitModule(M);
    visit(M.begin(), M.end());
  }

void llvm::InstVisitor< InstCombiner , Instruction * >::visit ( Function &  F )

inlineinherited

Definition at line 101 of file InstVisitor.h.

                          {
    static_cast<SubClass*>(this)->visitFunction(F);
    visit(F.begin(), F.end());
  }

void llvm::InstVisitor< InstCombiner , Instruction * >::visit ( BasicBlock &  BB )

inlineinherited

Definition at line 105 of file InstVisitor.h.

                             {
    static_cast<SubClass*>(this)->visitBasicBlock(BB);
    visit(BB.begin(), BB.end());
  }

void llvm::InstVisitor< InstCombiner , Instruction * >::visit ( Module *  M )

inlineinherited

Definition at line 111 of file InstVisitor.h.

{ visit(*M); }

void llvm::InstVisitor< InstCombiner , Instruction * >::visit ( Function *  F )

inlineinherited

Definition at line 112 of file InstVisitor.h.

{ visit(*F); }

void llvm::InstVisitor< InstCombiner , Instruction * >::visit ( BasicBlock *  BB )

inlineinherited

Definition at line 113 of file InstVisitor.h.

{ visit(*BB); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visit ( Instruction *  I )

inlineinherited

Definition at line 114 of file InstVisitor.h.

{ return visit(*I); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visit ( Instruction &  I )

inlineinherited

Definition at line 118 of file InstVisitor.h.

                              {
    switch (I.getOpcode()) {
    default: llvm_unreachable("Unknown instruction type encountered!");
      // Build the switch statement using the Instruction.def file...
#define HANDLE_INST(NUM, OPCODE, CLASS) \
    case Instruction::OPCODE: return \
           static_cast<SubClass*>(this)-> \
                      visit##OPCODE(static_cast<CLASS&>(I));
#include "llvm/IR/Instruction.def"
    }
  }

Instruction * InstCombiner::visitAdd

(

BinaryOperator &

I

)

Definition at line 1020 of file InstCombineAddSub.cpp.

                                                     {
   bool Changed = SimplifyAssociativeOrCommutative(I);
   Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);

   if (Value *V = SimplifyVectorOp(I))
     return ReplaceInstUsesWith(I, V);

   if (Value *V = SimplifyAddInst(LHS, RHS, I.hasNoSignedWrap(),
                                  I.hasNoUnsignedWrap(), DL))
     return ReplaceInstUsesWith(I, V);

   // (A*B)+(A*C) -> A*(B+C) etc
  if (Value *V = SimplifyUsingDistributiveLaws(I))
    return ReplaceInstUsesWith(I, V);

  if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
    // X + (signbit) --> X ^ signbit
    const APInt &Val = CI->getValue();
    if (Val.isSignBit())
      return BinaryOperator::CreateXor(LHS, RHS);

    // See if SimplifyDemandedBits can simplify this.  This handles stuff like
    // (X & 254)+1 -> (X&254)|1
    if (SimplifyDemandedInstructionBits(I))
      return &I;

    // zext(bool) + C -> bool ? C + 1 : C
    if (ZExtInst *ZI = dyn_cast<ZExtInst>(LHS))
      if (ZI->getSrcTy()->isIntegerTy(1))
        return SelectInst::Create(ZI->getOperand(0), AddOne(CI), CI);

    Value *XorLHS = nullptr; ConstantInt *XorRHS = nullptr;
    if (match(LHS, m_Xor(m_Value(XorLHS), m_ConstantInt(XorRHS)))) {
      uint32_t TySizeBits = I.getType()->getScalarSizeInBits();
      const APInt &RHSVal = CI->getValue();
      unsigned ExtendAmt = 0;
      // If we have ADD(XOR(AND(X, 0xFF), 0x80), 0xF..F80), it's a sext.
      // If we have ADD(XOR(AND(X, 0xFF), 0xF..F80), 0x80), it's a sext.
      if (XorRHS->getValue() == -RHSVal) {
        if (RHSVal.isPowerOf2())
          ExtendAmt = TySizeBits - RHSVal.logBase2() - 1;
        else if (XorRHS->getValue().isPowerOf2())
          ExtendAmt = TySizeBits - XorRHS->getValue().logBase2() - 1;
      }

      if (ExtendAmt) {
        APInt Mask = APInt::getHighBitsSet(TySizeBits, ExtendAmt);
        if (!MaskedValueIsZero(XorLHS, Mask))
          ExtendAmt = 0;
      }

      if (ExtendAmt) {
        Constant *ShAmt = ConstantInt::get(I.getType(), ExtendAmt);
        Value *NewShl = Builder->CreateShl(XorLHS, ShAmt, "sext");
        return BinaryOperator::CreateAShr(NewShl, ShAmt);
      }

      // If this is a xor that was canonicalized from a sub, turn it back into
      // a sub and fuse this add with it.
      if (LHS->hasOneUse() && (XorRHS->getValue()+1).isPowerOf2()) {
        IntegerType *IT = cast<IntegerType>(I.getType());
        APInt LHSKnownOne(IT->getBitWidth(), 0);
        APInt LHSKnownZero(IT->getBitWidth(), 0);
        computeKnownBits(XorLHS, LHSKnownZero, LHSKnownOne);
        if ((XorRHS->getValue() | LHSKnownZero).isAllOnesValue())
          return BinaryOperator::CreateSub(ConstantExpr::getAdd(XorRHS, CI),
                                           XorLHS);
      }
      // (X + signbit) + C could have gotten canonicalized to (X ^ signbit) + C,
      // transform them into (X + (signbit ^ C))
      if (XorRHS->getValue().isSignBit())
          return BinaryOperator::CreateAdd(XorLHS,
                                           ConstantExpr::getXor(XorRHS, CI));
    }
  }

  if (isa<Constant>(RHS) && isa<PHINode>(LHS))
    if (Instruction *NV = FoldOpIntoPhi(I))
      return NV;

  if (I.getType()->getScalarType()->isIntegerTy(1))
    return BinaryOperator::CreateXor(LHS, RHS);

  // X + X --> X << 1
  if (LHS == RHS) {
    BinaryOperator *New =
      BinaryOperator::CreateShl(LHS, ConstantInt::get(I.getType(), 1));
    New->setHasNoSignedWrap(I.hasNoSignedWrap());
    New->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
    return New;
  }

  // -A + B  -->  B - A
  // -A + -B  -->  -(A + B)
  if (Value *LHSV = dyn_castNegVal(LHS)) {
    if (!isa<Constant>(RHS))
      if (Value *RHSV = dyn_castNegVal(RHS)) {
        Value *NewAdd = Builder->CreateAdd(LHSV, RHSV, "sum");
        return BinaryOperator::CreateNeg(NewAdd);
      }

    return BinaryOperator::CreateSub(RHS, LHSV);
  }

  // A + -B  -->  A - B
  if (!isa<Constant>(RHS))
    if (Value *V = dyn_castNegVal(RHS))
      return BinaryOperator::CreateSub(LHS, V);

  if (Value *V = checkForNegativeOperand(I, Builder))
    return ReplaceInstUsesWith(I, V);

  // A+B --> A|B iff A and B have no bits set in common.
  if (IntegerType *IT = dyn_cast<IntegerType>(I.getType())) {
    APInt LHSKnownOne(IT->getBitWidth(), 0);
    APInt LHSKnownZero(IT->getBitWidth(), 0);
    computeKnownBits(LHS, LHSKnownZero, LHSKnownOne);
    if (LHSKnownZero != 0) {
      APInt RHSKnownOne(IT->getBitWidth(), 0);
      APInt RHSKnownZero(IT->getBitWidth(), 0);
      computeKnownBits(RHS, RHSKnownZero, RHSKnownOne);

      // No bits in common -> bitwise or.
      if ((LHSKnownZero|RHSKnownZero).isAllOnesValue())
        return BinaryOperator::CreateOr(LHS, RHS);
    }
  }

  if (Constant *CRHS = dyn_cast<Constant>(RHS)) {
    Value *X;
    if (match(LHS, m_Not(m_Value(X)))) // ~X + C --> (C-1) - X
      return BinaryOperator::CreateSub(SubOne(CRHS), X);
  }

  if (ConstantInt *CRHS = dyn_cast<ConstantInt>(RHS)) {
    // (X & FF00) + xx00  -> (X+xx00) & FF00
    Value *X;
    ConstantInt *C2;
    if (LHS->hasOneUse() &&
        match(LHS, m_And(m_Value(X), m_ConstantInt(C2))) &&
        CRHS->getValue() == (CRHS->getValue() & C2->getValue())) {
      // See if all bits from the first bit set in the Add RHS up are included
      // in the mask.  First, get the rightmost bit.
      const APInt &AddRHSV = CRHS->getValue();

      // Form a mask of all bits from the lowest bit added through the top.
      APInt AddRHSHighBits(~((AddRHSV & -AddRHSV)-1));

      // See if the and mask includes all of these bits.
      APInt AddRHSHighBitsAnd(AddRHSHighBits & C2->getValue());

      if (AddRHSHighBits == AddRHSHighBitsAnd) {
        // Okay, the xform is safe.  Insert the new add pronto.
        Value *NewAdd = Builder->CreateAdd(X, CRHS, LHS->getName());
        return BinaryOperator::CreateAnd(NewAdd, C2);
      }
    }

    // Try to fold constant add into select arguments.
    if (SelectInst *SI = dyn_cast<SelectInst>(LHS))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;
  }

  // add (select X 0 (sub n A)) A  -->  select X A n
  {
    SelectInst *SI = dyn_cast<SelectInst>(LHS);
    Value *A = RHS;
    if (!SI) {
      SI = dyn_cast<SelectInst>(RHS);
      A = LHS;
    }
    if (SI && SI->hasOneUse()) {
      Value *TV = SI->getTrueValue();
      Value *FV = SI->getFalseValue();
      Value *N;

      // Can we fold the add into the argument of the select?
      // We check both true and false select arguments for a matching subtract.
      if (match(FV, m_Zero()) && match(TV, m_Sub(m_Value(N), m_Specific(A))))
        // Fold the add into the true select value.
        return SelectInst::Create(SI->getCondition(), N, A);

      if (match(TV, m_Zero()) && match(FV, m_Sub(m_Value(N), m_Specific(A))))
        // Fold the add into the false select value.
        return SelectInst::Create(SI->getCondition(), A, N);
    }
  }

  // Check for (add (sext x), y), see if we can merge this into an
  // integer add followed by a sext.
  if (SExtInst *LHSConv = dyn_cast<SExtInst>(LHS)) {
    // (add (sext x), cst) --> (sext (add x, cst'))
    if (ConstantInt *RHSC = dyn_cast<ConstantInt>(RHS)) {
      Constant *CI =
        ConstantExpr::getTrunc(RHSC, LHSConv->getOperand(0)->getType());
      if (LHSConv->hasOneUse() &&
          ConstantExpr::getSExt(CI, I.getType()) == RHSC &&
          WillNotOverflowSignedAdd(LHSConv->getOperand(0), CI)) {
        // Insert the new, smaller add.
        Value *NewAdd = Builder->CreateNSWAdd(LHSConv->getOperand(0),
                                              CI, "addconv");
        return new SExtInst(NewAdd, I.getType());
      }
    }

    // (add (sext x), (sext y)) --> (sext (add int x, y))
    if (SExtInst *RHSConv = dyn_cast<SExtInst>(RHS)) {
      // Only do this if x/y have the same type, if at last one of them has a
      // single use (so we don't increase the number of sexts), and if the
      // integer add will not overflow.
      if (LHSConv->getOperand(0)->getType()==RHSConv->getOperand(0)->getType()&&
          (LHSConv->hasOneUse() || RHSConv->hasOneUse()) &&
          WillNotOverflowSignedAdd(LHSConv->getOperand(0),
                                   RHSConv->getOperand(0))) {
        // Insert the new integer add.
        Value *NewAdd = Builder->CreateNSWAdd(LHSConv->getOperand(0),
                                             RHSConv->getOperand(0), "addconv");
        return new SExtInst(NewAdd, I.getType());
      }
    }
  }

  // Check for (x & y) + (x ^ y)
  {
    Value *A = nullptr, *B = nullptr;
    if (match(RHS, m_Xor(m_Value(A), m_Value(B))) &&
        (match(LHS, m_And(m_Specific(A), m_Specific(B))) ||
         match(LHS, m_And(m_Specific(B), m_Specific(A)))))
      return BinaryOperator::CreateOr(A, B);

    if (match(LHS, m_Xor(m_Value(A), m_Value(B))) &&
        (match(RHS, m_And(m_Specific(A), m_Specific(B))) ||
         match(RHS, m_And(m_Specific(B), m_Specific(A)))))
      return BinaryOperator::CreateOr(A, B);
  }

  // TODO(jingyue): Consider WillNotOverflowSignedAdd and
  // WillNotOverflowUnsignedAdd to reduce the number of invocations of
  // computeKnownBits.
  if (!I.hasNoSignedWrap() && WillNotOverflowSignedAdd(LHS, RHS)) {
    Changed = true;
    I.setHasNoSignedWrap(true);
  }
  if (!I.hasNoUnsignedWrap() && WillNotOverflowUnsignedAdd(LHS, RHS)) {
    Changed = true;
    I.setHasNoUnsignedWrap(true);
  }

  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitAddrSpaceCast

(

AddrSpaceCastInst &

CI

)

Definition at line 1911 of file InstCombineCasts.cpp.

                                                                   {
  // If the destination pointer element type is not the same as the source's
  // first do a bitcast to the destination type, and then the addrspacecast.
  // This allows the cast to be exposed to other transforms.
  Value *Src = CI.getOperand(0);
  PointerType *SrcTy = cast<PointerType>(Src->getType()->getScalarType());
  PointerType *DestTy = cast<PointerType>(CI.getType()->getScalarType());

  Type *DestElemTy = DestTy->getElementType();
  if (SrcTy->getElementType() != DestElemTy) {
    Type *MidTy = PointerType::get(DestElemTy, SrcTy->getAddressSpace());
    if (VectorType *VT = dyn_cast<VectorType>(CI.getType())) {
      // Handle vectors of pointers.
      MidTy = VectorType::get(MidTy, VT->getNumElements());
    }

    Value *NewBitCast = Builder->CreateBitCast(Src, MidTy);
    return new AddrSpaceCastInst(NewBitCast, CI.getType());
  }

  return commonPointerCastTransforms(CI);
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitAddrSpaceCastInst ( AddrSpaceCastInst &  I )

inlineinherited

Definition at line 194 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitAllocaInst

(

AllocaInst &

AI

)

Definition at line 165 of file InstCombineLoadStoreAlloca.cpp.

                                                         {
  // Ensure that the alloca array size argument has type intptr_t, so that
  // any casting is exposed early.
  if (DL) {
    Type *IntPtrTy = DL->getIntPtrType(AI.getType());
    if (AI.getArraySize()->getType() != IntPtrTy) {
      Value *V = Builder->CreateIntCast(AI.getArraySize(),
                                        IntPtrTy, false);
      AI.setOperand(0, V);
      return &AI;
    }
  }

  // Convert: alloca Ty, C - where C is a constant != 1 into: alloca [C x Ty], 1
  if (AI.isArrayAllocation()) {  // Check C != 1
    if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
      Type *NewTy =
        ArrayType::get(AI.getAllocatedType(), C->getZExtValue());
      AllocaInst *New = Builder->CreateAlloca(NewTy, nullptr, AI.getName());
      New->setAlignment(AI.getAlignment());

      // Scan to the end of the allocation instructions, to skip over a block of
      // allocas if possible...also skip interleaved debug info
      //
      BasicBlock::iterator It = New;
      while (isa<AllocaInst>(*It) || isa<DbgInfoIntrinsic>(*It)) ++It;

      // Now that I is pointing to the first non-allocation-inst in the block,
      // insert our getelementptr instruction...
      //
      Type *IdxTy = DL
                  ? DL->getIntPtrType(AI.getType())
                  : Type::getInt64Ty(AI.getContext());
      Value *NullIdx = Constant::getNullValue(IdxTy);
      Value *Idx[2] = { NullIdx, NullIdx };
      Instruction *GEP =
        GetElementPtrInst::CreateInBounds(New, Idx, New->getName() + ".sub");
      InsertNewInstBefore(GEP, *It);

      // Now make everything use the getelementptr instead of the original
      // allocation.
      return ReplaceInstUsesWith(AI, GEP);
    } else if (isa<UndefValue>(AI.getArraySize())) {
      return ReplaceInstUsesWith(AI, Constant::getNullValue(AI.getType()));
    }
  }

  if (DL && AI.getAllocatedType()->isSized()) {
    // If the alignment is 0 (unspecified), assign it the preferred alignment.
    if (AI.getAlignment() == 0)
      AI.setAlignment(DL->getPrefTypeAlignment(AI.getAllocatedType()));

    // Move all alloca's of zero byte objects to the entry block and merge them
    // together.  Note that we only do this for alloca's, because malloc should
    // allocate and return a unique pointer, even for a zero byte allocation.
    if (DL->getTypeAllocSize(AI.getAllocatedType()) == 0) {
      // For a zero sized alloca there is no point in doing an array allocation.
      // This is helpful if the array size is a complicated expression not used
      // elsewhere.
      if (AI.isArrayAllocation()) {
        AI.setOperand(0, ConstantInt::get(AI.getArraySize()->getType(), 1));
        return &AI;
      }

      // Get the first instruction in the entry block.
      BasicBlock &EntryBlock = AI.getParent()->getParent()->getEntryBlock();
      Instruction *FirstInst = EntryBlock.getFirstNonPHIOrDbg();
      if (FirstInst != &AI) {
        // If the entry block doesn't start with a zero-size alloca then move
        // this one to the start of the entry block.  There is no problem with
        // dominance as the array size was forced to a constant earlier already.
        AllocaInst *EntryAI = dyn_cast<AllocaInst>(FirstInst);
        if (!EntryAI || !EntryAI->getAllocatedType()->isSized() ||
            DL->getTypeAllocSize(EntryAI->getAllocatedType()) != 0) {
          AI.moveBefore(FirstInst);
          return &AI;
        }

        // If the alignment of the entry block alloca is 0 (unspecified),
        // assign it the preferred alignment.
        if (EntryAI->getAlignment() == 0)
          EntryAI->setAlignment(
            DL->getPrefTypeAlignment(EntryAI->getAllocatedType()));
        // Replace this zero-sized alloca with the one at the start of the entry
        // block after ensuring that the address will be aligned enough for both
        // types.
        unsigned MaxAlign = std::max(EntryAI->getAlignment(),
                                     AI.getAlignment());
        EntryAI->setAlignment(MaxAlign);
        if (AI.getType() != EntryAI->getType())
          return new BitCastInst(EntryAI, AI.getType());
        return ReplaceInstUsesWith(AI, EntryAI);
      }
    }
  }

  if (AI.getAlignment()) {
    // Check to see if this allocation is only modified by a memcpy/memmove from
    // a constant global whose alignment is equal to or exceeds that of the
    // allocation.  If this is the case, we can change all users to use
    // the constant global instead.  This is commonly produced by the CFE by
    // constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
    // is only subsequently read.
    SmallVector<Instruction *, 4> ToDelete;
    if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
      unsigned SourceAlign = getOrEnforceKnownAlignment(Copy->getSource(),
                                                        AI.getAlignment(), DL);
      if (AI.getAlignment() <= SourceAlign) {
        DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
        DEBUG(dbgs() << "  memcpy = " << *Copy << '\n');
        for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
          EraseInstFromFunction(*ToDelete[i]);
        Constant *TheSrc = cast<Constant>(Copy->getSource());
        Constant *Cast
          = ConstantExpr::getPointerBitCastOrAddrSpaceCast(TheSrc, AI.getType());
        Instruction *NewI = ReplaceInstUsesWith(AI, Cast);
        EraseInstFromFunction(*Copy);
        ++NumGlobalCopies;
        return NewI;
      }
    }
  }

  // At last, use the generic allocation site handler to aggressively remove
  // unused allocas.
  return visitAllocSite(AI);
}

Instruction * InstCombiner::visitAllocSite

(

Instruction &

FI

)

Definition at line 1800 of file InstructionCombining.cpp.

                                                         {
  // If we have a malloc call which is only used in any amount of comparisons
  // to null and free calls, delete the calls and replace the comparisons with
  // true or false as appropriate.
  SmallVector<WeakVH, 64> Users;
  if (isAllocSiteRemovable(&MI, Users, TLI)) {
    for (unsigned i = 0, e = Users.size(); i != e; ++i) {
      Instruction *I = cast_or_null<Instruction>(&*Users[i]);
      if (!I) continue;

      if (ICmpInst *C = dyn_cast<ICmpInst>(I)) {
        ReplaceInstUsesWith(*C,
                            ConstantInt::get(Type::getInt1Ty(C->getContext()),
                                             C->isFalseWhenEqual()));
      } else if (isa<BitCastInst>(I) || isa<GetElementPtrInst>(I)) {
        ReplaceInstUsesWith(*I, UndefValue::get(I->getType()));
      } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
        if (II->getIntrinsicID() == Intrinsic::objectsize) {
          ConstantInt *CI = cast<ConstantInt>(II->getArgOperand(1));
          uint64_t DontKnow = CI->isZero() ? -1ULL : 0;
          ReplaceInstUsesWith(*I, ConstantInt::get(I->getType(), DontKnow));
        }
      }
      EraseInstFromFunction(*I);
    }

    if (InvokeInst *II = dyn_cast<InvokeInst>(&MI)) {
      // Replace invoke with a NOP intrinsic to maintain the original CFG
      Module *M = II->getParent()->getParent()->getParent();
      Function *F = Intrinsic::getDeclaration(M, Intrinsic::donothing);
      InvokeInst::Create(F, II->getNormalDest(), II->getUnwindDest(),
                         None, "", II->getParent());
    }
    return EraseInstFromFunction(MI);
  }
  return nullptr;
}

Instruction * InstCombiner::visitAnd

(

BinaryOperator &

I

)

Definition at line 1105 of file InstCombineAndOrXor.cpp.

                                                     {
  bool Changed = SimplifyAssociativeOrCommutative(I);
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyAndInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // (A|B)&(A|C) -> A|(B&C) etc
  if (Value *V = SimplifyUsingDistributiveLaws(I))
    return ReplaceInstUsesWith(I, V);

  // See if we can simplify any instructions used by the instruction whose sole
  // purpose is to compute bits we don't care about.
  if (SimplifyDemandedInstructionBits(I))
    return &I;

  if (ConstantInt *AndRHS = dyn_cast<ConstantInt>(Op1)) {
    const APInt &AndRHSMask = AndRHS->getValue();

    // Optimize a variety of ((val OP C1) & C2) combinations...
    if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0)) {
      Value *Op0LHS = Op0I->getOperand(0);
      Value *Op0RHS = Op0I->getOperand(1);
      switch (Op0I->getOpcode()) {
      default: break;
      case Instruction::Xor:
      case Instruction::Or: {
        // If the mask is only needed on one incoming arm, push it up.
        if (!Op0I->hasOneUse()) break;

        APInt NotAndRHS(~AndRHSMask);
        if (MaskedValueIsZero(Op0LHS, NotAndRHS)) {
          // Not masking anything out for the LHS, move to RHS.
          Value *NewRHS = Builder->CreateAnd(Op0RHS, AndRHS,
                                             Op0RHS->getName()+".masked");
          return BinaryOperator::Create(Op0I->getOpcode(), Op0LHS, NewRHS);
        }
        if (!isa<Constant>(Op0RHS) &&
            MaskedValueIsZero(Op0RHS, NotAndRHS)) {
          // Not masking anything out for the RHS, move to LHS.
          Value *NewLHS = Builder->CreateAnd(Op0LHS, AndRHS,
                                             Op0LHS->getName()+".masked");
          return BinaryOperator::Create(Op0I->getOpcode(), NewLHS, Op0RHS);
        }

        break;
      }
      case Instruction::Add:
        // ((A & N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == AndRHS.
        // ((A | N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == 0
        // ((A ^ N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == 0
        if (Value *V = FoldLogicalPlusAnd(Op0LHS, Op0RHS, AndRHS, false, I))
          return BinaryOperator::CreateAnd(V, AndRHS);
        if (Value *V = FoldLogicalPlusAnd(Op0RHS, Op0LHS, AndRHS, false, I))
          return BinaryOperator::CreateAnd(V, AndRHS);  // Add commutes
        break;

      case Instruction::Sub:
        // ((A & N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == AndRHS.
        // ((A | N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == 0
        // ((A ^ N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == 0
        if (Value *V = FoldLogicalPlusAnd(Op0LHS, Op0RHS, AndRHS, true, I))
          return BinaryOperator::CreateAnd(V, AndRHS);

        // (A - N) & AndRHS -> -N & AndRHS iff A&AndRHS==0 and AndRHS
        // has 1's for all bits that the subtraction with A might affect.
        if (Op0I->hasOneUse() && !match(Op0LHS, m_Zero())) {
          uint32_t BitWidth = AndRHSMask.getBitWidth();
          uint32_t Zeros = AndRHSMask.countLeadingZeros();
          APInt Mask = APInt::getLowBitsSet(BitWidth, BitWidth - Zeros);

          if (MaskedValueIsZero(Op0LHS, Mask)) {
            Value *NewNeg = Builder->CreateNeg(Op0RHS);
            return BinaryOperator::CreateAnd(NewNeg, AndRHS);
          }
        }
        break;

      case Instruction::Shl:
      case Instruction::LShr:
        // (1 << x) & 1 --> zext(x == 0)
        // (1 >> x) & 1 --> zext(x == 0)
        if (AndRHSMask == 1 && Op0LHS == AndRHS) {
          Value *NewICmp =
            Builder->CreateICmpEQ(Op0RHS, Constant::getNullValue(I.getType()));
          return new ZExtInst(NewICmp, I.getType());
        }
        break;
      }

      if (ConstantInt *Op0CI = dyn_cast<ConstantInt>(Op0I->getOperand(1)))
        if (Instruction *Res = OptAndOp(Op0I, Op0CI, AndRHS, I))
          return Res;
    }

    // If this is an integer truncation, and if the source is an 'and' with
    // immediate, transform it.  This frequently occurs for bitfield accesses.
    {
      Value *X = nullptr; ConstantInt *YC = nullptr;
      if (match(Op0, m_Trunc(m_And(m_Value(X), m_ConstantInt(YC))))) {
        // Change: and (trunc (and X, YC) to T), C2
        // into  : and (trunc X to T), trunc(YC) & C2
        // This will fold the two constants together, which may allow
        // other simplifications.
        Value *NewCast = Builder->CreateTrunc(X, I.getType(), "and.shrunk");
        Constant *C3 = ConstantExpr::getTrunc(YC, I.getType());
        C3 = ConstantExpr::getAnd(C3, AndRHS);
        return BinaryOperator::CreateAnd(NewCast, C3);
      }
    }

    // Try to fold constant and into select arguments.
    if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;
    if (isa<PHINode>(Op0))
      if (Instruction *NV = FoldOpIntoPhi(I))
        return NV;
  }


  // (~A & ~B) == (~(A | B)) - De Morgan's Law
  if (Value *Op0NotVal = dyn_castNotVal(Op0))
    if (Value *Op1NotVal = dyn_castNotVal(Op1))
      if (Op0->hasOneUse() && Op1->hasOneUse()) {
        Value *Or = Builder->CreateOr(Op0NotVal, Op1NotVal,
                                      I.getName()+".demorgan");
        return BinaryOperator::CreateNot(Or);
      }

  {
    Value *A = nullptr, *B = nullptr, *C = nullptr, *D = nullptr;
    // (A|B) & ~(A&B) -> A^B
    if (match(Op0, m_Or(m_Value(A), m_Value(B))) &&
        match(Op1, m_Not(m_And(m_Value(C), m_Value(D)))) &&
        ((A == C && B == D) || (A == D && B == C)))
      return BinaryOperator::CreateXor(A, B);

    // ~(A&B) & (A|B) -> A^B
    if (match(Op1, m_Or(m_Value(A), m_Value(B))) &&
        match(Op0, m_Not(m_And(m_Value(C), m_Value(D)))) &&
        ((A == C && B == D) || (A == D && B == C)))
      return BinaryOperator::CreateXor(A, B);

    // A&(A^B) => A & ~B
    {
      Value *tmpOp0 = Op0;
      Value *tmpOp1 = Op1;
      if (Op0->hasOneUse() &&
          match(Op0, m_Xor(m_Value(A), m_Value(B)))) {
        if (A == Op1 || B == Op1 ) {
          tmpOp1 = Op0;
          tmpOp0 = Op1;
          // Simplify below
        }
      }

      if (tmpOp1->hasOneUse() &&
          match(tmpOp1, m_Xor(m_Value(A), m_Value(B)))) {
        if (B == tmpOp0) {
          std::swap(A, B);
        }
        // Notice that the patten (A&(~B)) is actually (A&(-1^B)), so if
        // A is originally -1 (or a vector of -1 and undefs), then we enter
        // an endless loop. By checking that A is non-constant we ensure that
        // we will never get to the loop.
        if (A == tmpOp0 && !isa<Constant>(A)) // A&(A^B) -> A & ~B
          return BinaryOperator::CreateAnd(A, Builder->CreateNot(B));
      }
    }

    // (A&((~A)|B)) -> A&B
    if (match(Op0, m_Or(m_Not(m_Specific(Op1)), m_Value(A))) ||
        match(Op0, m_Or(m_Value(A), m_Not(m_Specific(Op1)))))
      return BinaryOperator::CreateAnd(A, Op1);
    if (match(Op1, m_Or(m_Not(m_Specific(Op0)), m_Value(A))) ||
        match(Op1, m_Or(m_Value(A), m_Not(m_Specific(Op0)))))
      return BinaryOperator::CreateAnd(A, Op0);
  }

  if (ICmpInst *RHS = dyn_cast<ICmpInst>(Op1))
    if (ICmpInst *LHS = dyn_cast<ICmpInst>(Op0))
      if (Value *Res = FoldAndOfICmps(LHS, RHS))
        return ReplaceInstUsesWith(I, Res);

  // If and'ing two fcmp, try combine them into one.
  if (FCmpInst *LHS = dyn_cast<FCmpInst>(I.getOperand(0)))
    if (FCmpInst *RHS = dyn_cast<FCmpInst>(I.getOperand(1)))
      if (Value *Res = FoldAndOfFCmps(LHS, RHS))
        return ReplaceInstUsesWith(I, Res);


  // fold (and (cast A), (cast B)) -> (cast (and A, B))
  if (CastInst *Op0C = dyn_cast<CastInst>(Op0))
    if (CastInst *Op1C = dyn_cast<CastInst>(Op1)) {
      Type *SrcTy = Op0C->getOperand(0)->getType();
      if (Op0C->getOpcode() == Op1C->getOpcode() && // same cast kind ?
          SrcTy == Op1C->getOperand(0)->getType() &&
          SrcTy->isIntOrIntVectorTy()) {
        Value *Op0COp = Op0C->getOperand(0), *Op1COp = Op1C->getOperand(0);

        // Only do this if the casts both really cause code to be generated.
        if (ShouldOptimizeCast(Op0C->getOpcode(), Op0COp, I.getType()) &&
            ShouldOptimizeCast(Op1C->getOpcode(), Op1COp, I.getType())) {
          Value *NewOp = Builder->CreateAnd(Op0COp, Op1COp, I.getName());
          return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
        }

        // If this is and(cast(icmp), cast(icmp)), try to fold this even if the
        // cast is otherwise not optimizable.  This happens for vector sexts.
        if (ICmpInst *RHS = dyn_cast<ICmpInst>(Op1COp))
          if (ICmpInst *LHS = dyn_cast<ICmpInst>(Op0COp))
            if (Value *Res = FoldAndOfICmps(LHS, RHS))
              return CastInst::Create(Op0C->getOpcode(), Res, I.getType());

        // If this is and(cast(fcmp), cast(fcmp)), try to fold this even if the
        // cast is otherwise not optimizable.  This happens for vector sexts.
        if (FCmpInst *RHS = dyn_cast<FCmpInst>(Op1COp))
          if (FCmpInst *LHS = dyn_cast<FCmpInst>(Op0COp))
            if (Value *Res = FoldAndOfFCmps(LHS, RHS))
              return CastInst::Create(Op0C->getOpcode(), Res, I.getType());
      }
    }

  // (X >> Z) & (Y >> Z)  -> (X&Y) >> Z  for all shifts.
  if (BinaryOperator *SI1 = dyn_cast<BinaryOperator>(Op1)) {
    if (BinaryOperator *SI0 = dyn_cast<BinaryOperator>(Op0))
      if (SI0->isShift() && SI0->getOpcode() == SI1->getOpcode() &&
          SI0->getOperand(1) == SI1->getOperand(1) &&
          (SI0->hasOneUse() || SI1->hasOneUse())) {
        Value *NewOp =
          Builder->CreateAnd(SI0->getOperand(0), SI1->getOperand(0),
                             SI0->getName());
        return BinaryOperator::Create(SI1->getOpcode(), NewOp,
                                      SI1->getOperand(1));
      }
  }

  {
    Value *X = nullptr;
    bool OpsSwapped = false;
    // Canonicalize SExt or Not to the LHS
    if (match(Op1, m_SExt(m_Value())) ||
        match(Op1, m_Not(m_Value()))) {
      std::swap(Op0, Op1);
      OpsSwapped = true;
    }

    // Fold (and (sext bool to A), B) --> (select bool, B, 0)
    if (match(Op0, m_SExt(m_Value(X))) &&
        X->getType()->getScalarType()->isIntegerTy(1)) {
      Value *Zero = Constant::getNullValue(Op1->getType());
      return SelectInst::Create(X, Op1, Zero);
    }

    // Fold (and ~(sext bool to A), B) --> (select bool, 0, B)
    if (match(Op0, m_Not(m_SExt(m_Value(X)))) &&
        X->getType()->getScalarType()->isIntegerTy(1)) {
      Value *Zero = Constant::getNullValue(Op0->getType());
      return SelectInst::Create(X, Zero, Op1);
    }

    if (OpsSwapped)
      std::swap(Op0, Op1);
  }

  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitAShr

(

BinaryOperator &

I

)

Definition at line 772 of file InstCombineShifts.cpp.

                                                      {
  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyAShrInst(I.getOperand(0), I.getOperand(1),
                                  I.isExact(), DL))
    return ReplaceInstUsesWith(I, V);

  if (Instruction *R = commonShiftTransforms(I))
    return R;

  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (ConstantInt *Op1C = dyn_cast<ConstantInt>(Op1)) {
    unsigned ShAmt = Op1C->getZExtValue();

    // If the input is a SHL by the same constant (ashr (shl X, C), C), then we
    // have a sign-extend idiom.
    Value *X;
    if (match(Op0, m_Shl(m_Value(X), m_Specific(Op1)))) {
      // If the input is an extension from the shifted amount value, e.g.
      //   %x = zext i8 %A to i32
      //   %y = shl i32 %x, 24
      //   %z = ashr %y, 24
      // then turn this into "z = sext i8 A to i32".
      if (ZExtInst *ZI = dyn_cast<ZExtInst>(X)) {
        uint32_t SrcBits = ZI->getOperand(0)->getType()->getScalarSizeInBits();
        uint32_t DestBits = ZI->getType()->getScalarSizeInBits();
        if (Op1C->getZExtValue() == DestBits-SrcBits)
          return new SExtInst(ZI->getOperand(0), ZI->getType());
      }
    }

    // If the shifted-out value is known-zero, then this is an exact shift.
    if (!I.isExact() &&
        MaskedValueIsZero(Op0,APInt::getLowBitsSet(Op1C->getBitWidth(),ShAmt))){
      I.setIsExact();
      return &I;
    }
  }

  // See if we can turn a signed shr into an unsigned shr.
  if (MaskedValueIsZero(Op0,
                        APInt::getSignBit(I.getType()->getScalarSizeInBits())))
    return BinaryOperator::CreateLShr(Op0, Op1);

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitAtomicCmpXchgInst ( AtomicCmpXchgInst &  I )

inlineinherited

Definition at line 177 of file InstVisitor.h.

{ DELEGATE(Instruction);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitAtomicRMWInst ( AtomicRMWInst &  I )

inlineinherited

Definition at line 178 of file InstVisitor.h.

{ DELEGATE(Instruction);}

void llvm::InstVisitor< InstCombiner , Instruction * >::visitBasicBlock ( BasicBlock &  BB )

inlineinherited

Definition at line 143 of file InstVisitor.h.

{}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitBinaryOperator ( BinaryOperator &  I )

inlineinherited

Definition at line 232 of file InstVisitor.h.

{ DELEGATE(Instruction);}

Instruction * InstCombiner::visitBitCast

(

BitCastInst &

CI

)

Definition at line 1783 of file InstCombineCasts.cpp.

                                                       {
  // If the operands are integer typed then apply the integer transforms,
  // otherwise just apply the common ones.
  Value *Src = CI.getOperand(0);
  Type *SrcTy = Src->getType();
  Type *DestTy = CI.getType();

  // Get rid of casts from one type to the same type. These are useless and can
  // be replaced by the operand.
  if (DestTy == Src->getType())
    return ReplaceInstUsesWith(CI, Src);

  if (PointerType *DstPTy = dyn_cast<PointerType>(DestTy)) {
    PointerType *SrcPTy = cast<PointerType>(SrcTy);
    Type *DstElTy = DstPTy->getElementType();
    Type *SrcElTy = SrcPTy->getElementType();

    // If we are casting a alloca to a pointer to a type of the same
    // size, rewrite the allocation instruction to allocate the "right" type.
    // There is no need to modify malloc calls because it is their bitcast that
    // needs to be cleaned up.
    if (AllocaInst *AI = dyn_cast<AllocaInst>(Src))
      if (Instruction *V = PromoteCastOfAllocation(CI, *AI))
        return V;

    // If the source and destination are pointers, and this cast is equivalent
    // to a getelementptr X, 0, 0, 0...  turn it into the appropriate gep.
    // This can enhance SROA and other transforms that want type-safe pointers.
    Constant *ZeroUInt =
      Constant::getNullValue(Type::getInt32Ty(CI.getContext()));
    unsigned NumZeros = 0;
    while (SrcElTy != DstElTy &&
           isa<CompositeType>(SrcElTy) && !SrcElTy->isPointerTy() &&
           SrcElTy->getNumContainedTypes() /* not "{}" */) {
      SrcElTy = cast<CompositeType>(SrcElTy)->getTypeAtIndex(ZeroUInt);
      ++NumZeros;
    }

    // If we found a path from the src to dest, create the getelementptr now.
    if (SrcElTy == DstElTy) {
      SmallVector<Value*, 8> Idxs(NumZeros+1, ZeroUInt);
      return GetElementPtrInst::CreateInBounds(Src, Idxs);
    }
  }

  // Try to optimize int -> float bitcasts.
  if ((DestTy->isFloatTy() || DestTy->isDoubleTy()) && isa<IntegerType>(SrcTy))
    if (Instruction *I = OptimizeIntToFloatBitCast(CI, *this))
      return I;

  if (VectorType *DestVTy = dyn_cast<VectorType>(DestTy)) {
    if (DestVTy->getNumElements() == 1 && !SrcTy->isVectorTy()) {
      Value *Elem = Builder->CreateBitCast(Src, DestVTy->getElementType());
      return InsertElementInst::Create(UndefValue::get(DestTy), Elem,
                     Constant::getNullValue(Type::getInt32Ty(CI.getContext())));
      // FIXME: Canonicalize bitcast(insertelement) -> insertelement(bitcast)
    }

    if (isa<IntegerType>(SrcTy)) {
      // If this is a cast from an integer to vector, check to see if the input
      // is a trunc or zext of a bitcast from vector.  If so, we can replace all
      // the casts with a shuffle and (potentially) a bitcast.
      if (isa<TruncInst>(Src) || isa<ZExtInst>(Src)) {
        CastInst *SrcCast = cast<CastInst>(Src);
        if (BitCastInst *BCIn = dyn_cast<BitCastInst>(SrcCast->getOperand(0)))
          if (isa<VectorType>(BCIn->getOperand(0)->getType()))
            if (Instruction *I = OptimizeVectorResize(BCIn->getOperand(0),
                                               cast<VectorType>(DestTy), *this))
              return I;
      }

      // If the input is an 'or' instruction, we may be doing shifts and ors to
      // assemble the elements of the vector manually.  Try to rip the code out
      // and replace it with insertelements.
      if (Value *V = OptimizeIntegerToVectorInsertions(CI, *this))
        return ReplaceInstUsesWith(CI, V);
    }
  }

  if (VectorType *SrcVTy = dyn_cast<VectorType>(SrcTy)) {
    if (SrcVTy->getNumElements() == 1) {
      // If our destination is not a vector, then make this a straight
      // scalar-scalar cast.
      if (!DestTy->isVectorTy()) {
        Value *Elem =
          Builder->CreateExtractElement(Src,
                     Constant::getNullValue(Type::getInt32Ty(CI.getContext())));
        return CastInst::Create(Instruction::BitCast, Elem, DestTy);
      }

      // Otherwise, see if our source is an insert. If so, then use the scalar
      // component directly.
      if (InsertElementInst *IEI =
            dyn_cast<InsertElementInst>(CI.getOperand(0)))
        return CastInst::Create(Instruction::BitCast, IEI->getOperand(1),
                                DestTy);
    }
  }

  if (ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(Src)) {
    // Okay, we have (bitcast (shuffle ..)).  Check to see if this is
    // a bitcast to a vector with the same # elts.
    if (SVI->hasOneUse() && DestTy->isVectorTy() &&
        DestTy->getVectorNumElements() == SVI->getType()->getNumElements() &&
        SVI->getType()->getNumElements() ==
        SVI->getOperand(0)->getType()->getVectorNumElements()) {
      BitCastInst *Tmp;
      // If either of the operands is a cast from CI.getType(), then
      // evaluating the shuffle in the casted destination's type will allow
      // us to eliminate at least one cast.
      if (((Tmp = dyn_cast<BitCastInst>(SVI->getOperand(0))) &&
           Tmp->getOperand(0)->getType() == DestTy) ||
          ((Tmp = dyn_cast<BitCastInst>(SVI->getOperand(1))) &&
           Tmp->getOperand(0)->getType() == DestTy)) {
        Value *LHS = Builder->CreateBitCast(SVI->getOperand(0), DestTy);
        Value *RHS = Builder->CreateBitCast(SVI->getOperand(1), DestTy);
        // Return a new shuffle vector.  Use the same element ID's, as we
        // know the vector types match #elts.
        return new ShuffleVectorInst(LHS, RHS, SVI->getOperand(2));
      }
    }
  }

  if (SrcTy->isPointerTy())
    return commonPointerCastTransforms(CI);
  return commonCastTransforms(CI);
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitBitCastInst ( BitCastInst &  I )

inlineinherited

Definition at line 193 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitBranchInst

(

BranchInst &

BI

)

Definition at line 1927 of file InstructionCombining.cpp.

                                                         {
  // Change br (not X), label True, label False to: br X, label False, True
  Value *X = nullptr;
  BasicBlock *TrueDest;
  BasicBlock *FalseDest;
  if (match(&BI, m_Br(m_Not(m_Value(X)), TrueDest, FalseDest)) &&
      !isa<Constant>(X)) {
    // Swap Destinations and condition...
    BI.setCondition(X);
    BI.swapSuccessors();
    return &BI;
  }

  // Canonicalize fcmp_one -> fcmp_oeq
  FCmpInst::Predicate FPred; Value *Y;
  if (match(&BI, m_Br(m_FCmp(FPred, m_Value(X), m_Value(Y)),
                             TrueDest, FalseDest)) &&
      BI.getCondition()->hasOneUse())
    if (FPred == FCmpInst::FCMP_ONE || FPred == FCmpInst::FCMP_OLE ||
        FPred == FCmpInst::FCMP_OGE) {
      FCmpInst *Cond = cast<FCmpInst>(BI.getCondition());
      Cond->setPredicate(FCmpInst::getInversePredicate(FPred));

      // Swap Destinations and condition.
      BI.swapSuccessors();
      Worklist.Add(Cond);
      return &BI;
    }

  // Canonicalize icmp_ne -> icmp_eq
  ICmpInst::Predicate IPred;
  if (match(&BI, m_Br(m_ICmp(IPred, m_Value(X), m_Value(Y)),
                      TrueDest, FalseDest)) &&
      BI.getCondition()->hasOneUse())
    if (IPred == ICmpInst::ICMP_NE  || IPred == ICmpInst::ICMP_ULE ||
        IPred == ICmpInst::ICMP_SLE || IPred == ICmpInst::ICMP_UGE ||
        IPred == ICmpInst::ICMP_SGE) {
      ICmpInst *Cond = cast<ICmpInst>(BI.getCondition());
      Cond->setPredicate(ICmpInst::getInversePredicate(IPred));
      // Swap Destinations and condition.
      BI.swapSuccessors();
      Worklist.Add(Cond);
      return &BI;
    }

  return nullptr;
}

Instruction * InstCombiner::visitCallInst

(

CallInst &

CI

)

visitCallInst - CallInst simplification. This mostly only handles folding of intrinsic instructions. For normal calls, it allows visitCallSite to do the heavy lifting.

Definition at line 203 of file InstCombineCalls.cpp.

                                                     {
  if (isFreeCall(&CI, TLI))
    return visitFree(CI);

  // If the caller function is nounwind, mark the call as nounwind, even if the
  // callee isn't.
  if (CI.getParent()->getParent()->doesNotThrow() &&
      !CI.doesNotThrow()) {
    CI.setDoesNotThrow();
    return &CI;
  }

  IntrinsicInst *II = dyn_cast<IntrinsicInst>(&CI);
  if (!II) return visitCallSite(&CI);

  // Intrinsics cannot occur in an invoke, so handle them here instead of in
  // visitCallSite.
  if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(II)) {
    bool Changed = false;

    // memmove/cpy/set of zero bytes is a noop.
    if (Constant *NumBytes = dyn_cast<Constant>(MI->getLength())) {
      if (NumBytes->isNullValue())
        return EraseInstFromFunction(CI);

      if (ConstantInt *CI = dyn_cast<ConstantInt>(NumBytes))
        if (CI->getZExtValue() == 1) {
          // Replace the instruction with just byte operations.  We would
          // transform other cases to loads/stores, but we don't know if
          // alignment is sufficient.
        }
    }

    // No other transformations apply to volatile transfers.
    if (MI->isVolatile())
      return nullptr;

    // If we have a memmove and the source operation is a constant global,
    // then the source and dest pointers can't alias, so we can change this
    // into a call to memcpy.
    if (MemMoveInst *MMI = dyn_cast<MemMoveInst>(MI)) {
      if (GlobalVariable *GVSrc = dyn_cast<GlobalVariable>(MMI->getSource()))
        if (GVSrc->isConstant()) {
          Module *M = CI.getParent()->getParent()->getParent();
          Intrinsic::ID MemCpyID = Intrinsic::memcpy;
          Type *Tys[3] = { CI.getArgOperand(0)->getType(),
                           CI.getArgOperand(1)->getType(),
                           CI.getArgOperand(2)->getType() };
          CI.setCalledFunction(Intrinsic::getDeclaration(M, MemCpyID, Tys));
          Changed = true;
        }
    }

    if (MemTransferInst *MTI = dyn_cast<MemTransferInst>(MI)) {
      // memmove(x,x,size) -> noop.
      if (MTI->getSource() == MTI->getDest())
        return EraseInstFromFunction(CI);
    }

    // If we can determine a pointer alignment that is bigger than currently
    // set, update the alignment.
    if (isa<MemTransferInst>(MI)) {
      if (Instruction *I = SimplifyMemTransfer(MI))
        return I;
    } else if (MemSetInst *MSI = dyn_cast<MemSetInst>(MI)) {
      if (Instruction *I = SimplifyMemSet(MSI))
        return I;
    }

    if (Changed) return II;
  }

  switch (II->getIntrinsicID()) {
  default: break;
  case Intrinsic::objectsize: {
    uint64_t Size;
    if (getObjectSize(II->getArgOperand(0), Size, DL, TLI))
      return ReplaceInstUsesWith(CI, ConstantInt::get(CI.getType(), Size));
    return nullptr;
  }
  case Intrinsic::bswap: {
    Value *IIOperand = II->getArgOperand(0);
    Value *X = nullptr;

    // bswap(bswap(x)) -> x
    if (match(IIOperand, m_BSwap(m_Value(X))))
        return ReplaceInstUsesWith(CI, X);

    // bswap(trunc(bswap(x))) -> trunc(lshr(x, c))
    if (match(IIOperand, m_Trunc(m_BSwap(m_Value(X))))) {
      unsigned C = X->getType()->getPrimitiveSizeInBits() -
        IIOperand->getType()->getPrimitiveSizeInBits();
      Value *CV = ConstantInt::get(X->getType(), C);
      Value *V = Builder->CreateLShr(X, CV);
      return new TruncInst(V, IIOperand->getType());
    }
    break;
  }

  case Intrinsic::powi:
    if (ConstantInt *Power = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
      // powi(x, 0) -> 1.0
      if (Power->isZero())
        return ReplaceInstUsesWith(CI, ConstantFP::get(CI.getType(), 1.0));
      // powi(x, 1) -> x
      if (Power->isOne())
        return ReplaceInstUsesWith(CI, II->getArgOperand(0));
      // powi(x, -1) -> 1/x
      if (Power->isAllOnesValue())
        return BinaryOperator::CreateFDiv(ConstantFP::get(CI.getType(), 1.0),
                                          II->getArgOperand(0));
    }
    break;
  case Intrinsic::cttz: {
    // If all bits below the first known one are known zero,
    // this value is constant.
    IntegerType *IT = dyn_cast<IntegerType>(II->getArgOperand(0)->getType());
    // FIXME: Try to simplify vectors of integers.
    if (!IT) break;
    uint32_t BitWidth = IT->getBitWidth();
    APInt KnownZero(BitWidth, 0);
    APInt KnownOne(BitWidth, 0);
    computeKnownBits(II->getArgOperand(0), KnownZero, KnownOne);
    unsigned TrailingZeros = KnownOne.countTrailingZeros();
    APInt Mask(APInt::getLowBitsSet(BitWidth, TrailingZeros));
    if ((Mask & KnownZero) == Mask)
      return ReplaceInstUsesWith(CI, ConstantInt::get(IT,
                                 APInt(BitWidth, TrailingZeros)));

    }
    break;
  case Intrinsic::ctlz: {
    // If all bits above the first known one are known zero,
    // this value is constant.
    IntegerType *IT = dyn_cast<IntegerType>(II->getArgOperand(0)->getType());
    // FIXME: Try to simplify vectors of integers.
    if (!IT) break;
    uint32_t BitWidth = IT->getBitWidth();
    APInt KnownZero(BitWidth, 0);
    APInt KnownOne(BitWidth, 0);
    computeKnownBits(II->getArgOperand(0), KnownZero, KnownOne);
    unsigned LeadingZeros = KnownOne.countLeadingZeros();
    APInt Mask(APInt::getHighBitsSet(BitWidth, LeadingZeros));
    if ((Mask & KnownZero) == Mask)
      return ReplaceInstUsesWith(CI, ConstantInt::get(IT,
                                 APInt(BitWidth, LeadingZeros)));

    }
    break;
  case Intrinsic::uadd_with_overflow: {
    Value *LHS = II->getArgOperand(0), *RHS = II->getArgOperand(1);
    IntegerType *IT = cast<IntegerType>(II->getArgOperand(0)->getType());
    uint32_t BitWidth = IT->getBitWidth();
    APInt LHSKnownZero(BitWidth, 0);
    APInt LHSKnownOne(BitWidth, 0);
    computeKnownBits(LHS, LHSKnownZero, LHSKnownOne);
    bool LHSKnownNegative = LHSKnownOne[BitWidth - 1];
    bool LHSKnownPositive = LHSKnownZero[BitWidth - 1];

    if (LHSKnownNegative || LHSKnownPositive) {
      APInt RHSKnownZero(BitWidth, 0);
      APInt RHSKnownOne(BitWidth, 0);
      computeKnownBits(RHS, RHSKnownZero, RHSKnownOne);
      bool RHSKnownNegative = RHSKnownOne[BitWidth - 1];
      bool RHSKnownPositive = RHSKnownZero[BitWidth - 1];
      if (LHSKnownNegative && RHSKnownNegative) {
        // The sign bit is set in both cases: this MUST overflow.
        // Create a simple add instruction, and insert it into the struct.
        Value *Add = Builder->CreateAdd(LHS, RHS);
        Add->takeName(&CI);
        Constant *V[] = {
          UndefValue::get(LHS->getType()),
          ConstantInt::getTrue(II->getContext())
        };
        StructType *ST = cast<StructType>(II->getType());
        Constant *Struct = ConstantStruct::get(ST, V);
        return InsertValueInst::Create(Struct, Add, 0);
      }

      if (LHSKnownPositive && RHSKnownPositive) {
        // The sign bit is clear in both cases: this CANNOT overflow.
        // Create a simple add instruction, and insert it into the struct.
        Value *Add = Builder->CreateNUWAdd(LHS, RHS);
        Add->takeName(&CI);
        Constant *V[] = {
          UndefValue::get(LHS->getType()),
          ConstantInt::getFalse(II->getContext())
        };
        StructType *ST = cast<StructType>(II->getType());
        Constant *Struct = ConstantStruct::get(ST, V);
        return InsertValueInst::Create(Struct, Add, 0);
      }
    }
  }
  // FALL THROUGH uadd into sadd
  case Intrinsic::sadd_with_overflow:
    // Canonicalize constants into the RHS.
    if (isa<Constant>(II->getArgOperand(0)) &&
        !isa<Constant>(II->getArgOperand(1))) {
      Value *LHS = II->getArgOperand(0);
      II->setArgOperand(0, II->getArgOperand(1));
      II->setArgOperand(1, LHS);
      return II;
    }

    // X + undef -> undef
    if (isa<UndefValue>(II->getArgOperand(1)))
      return ReplaceInstUsesWith(CI, UndefValue::get(II->getType()));

    if (ConstantInt *RHS = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
      // X + 0 -> {X, false}
      if (RHS->isZero()) {
        Constant *V[] = {
          UndefValue::get(II->getArgOperand(0)->getType()),
          ConstantInt::getFalse(II->getContext())
        };
        Constant *Struct =
          ConstantStruct::get(cast<StructType>(II->getType()), V);
        return InsertValueInst::Create(Struct, II->getArgOperand(0), 0);
      }
    }

    // We can strength reduce reduce this signed add into a regular add if we
    // can prove that it will never overflow.
    if (II->getIntrinsicID() == Intrinsic::sadd_with_overflow) {
      Value *LHS = II->getArgOperand(0), *RHS = II->getArgOperand(1);
      if (WillNotOverflowSignedAdd(LHS, RHS)) {
        Value *Add = Builder->CreateNSWAdd(LHS, RHS);
        Add->takeName(&CI);
        Constant *V[] = {UndefValue::get(Add->getType()), Builder->getFalse()};
        StructType *ST = cast<StructType>(II->getType());
        Constant *Struct = ConstantStruct::get(ST, V);
        return InsertValueInst::Create(Struct, Add, 0);
      }
    }

    break;
  case Intrinsic::usub_with_overflow:
  case Intrinsic::ssub_with_overflow:
    // undef - X -> undef
    // X - undef -> undef
    if (isa<UndefValue>(II->getArgOperand(0)) ||
        isa<UndefValue>(II->getArgOperand(1)))
      return ReplaceInstUsesWith(CI, UndefValue::get(II->getType()));

    if (ConstantInt *RHS = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
      // X - 0 -> {X, false}
      if (RHS->isZero()) {
        Constant *V[] = {
          UndefValue::get(II->getArgOperand(0)->getType()),
          ConstantInt::getFalse(II->getContext())
        };
        Constant *Struct =
          ConstantStruct::get(cast<StructType>(II->getType()), V);
        return InsertValueInst::Create(Struct, II->getArgOperand(0), 0);
      }
    }
    break;
  case Intrinsic::umul_with_overflow: {
    Value *LHS = II->getArgOperand(0), *RHS = II->getArgOperand(1);
    unsigned BitWidth = cast<IntegerType>(LHS->getType())->getBitWidth();

    APInt LHSKnownZero(BitWidth, 0);
    APInt LHSKnownOne(BitWidth, 0);
    computeKnownBits(LHS, LHSKnownZero, LHSKnownOne);
    APInt RHSKnownZero(BitWidth, 0);
    APInt RHSKnownOne(BitWidth, 0);
    computeKnownBits(RHS, RHSKnownZero, RHSKnownOne);

    // Get the largest possible values for each operand.
    APInt LHSMax = ~LHSKnownZero;
    APInt RHSMax = ~RHSKnownZero;

    // If multiplying the maximum values does not overflow then we can turn
    // this into a plain NUW mul.
    bool Overflow;
    LHSMax.umul_ov(RHSMax, Overflow);
    if (!Overflow) {
      Value *Mul = Builder->CreateNUWMul(LHS, RHS, "umul_with_overflow");
      Constant *V[] = {
        UndefValue::get(LHS->getType()),
        Builder->getFalse()
      };
      Constant *Struct = ConstantStruct::get(cast<StructType>(II->getType()),V);
      return InsertValueInst::Create(Struct, Mul, 0);
    }
  } // FALL THROUGH
  case Intrinsic::smul_with_overflow:
    // Canonicalize constants into the RHS.
    if (isa<Constant>(II->getArgOperand(0)) &&
        !isa<Constant>(II->getArgOperand(1))) {
      Value *LHS = II->getArgOperand(0);
      II->setArgOperand(0, II->getArgOperand(1));
      II->setArgOperand(1, LHS);
      return II;
    }

    // X * undef -> undef
    if (isa<UndefValue>(II->getArgOperand(1)))
      return ReplaceInstUsesWith(CI, UndefValue::get(II->getType()));

    if (ConstantInt *RHSI = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
      // X*0 -> {0, false}
      if (RHSI->isZero())
        return ReplaceInstUsesWith(CI, Constant::getNullValue(II->getType()));

      // X * 1 -> {X, false}
      if (RHSI->equalsInt(1)) {
        Constant *V[] = {
          UndefValue::get(II->getArgOperand(0)->getType()),
          ConstantInt::getFalse(II->getContext())
        };
        Constant *Struct =
          ConstantStruct::get(cast<StructType>(II->getType()), V);
        return InsertValueInst::Create(Struct, II->getArgOperand(0), 0);
      }
    }
    break;
  case Intrinsic::ppc_altivec_lvx:
  case Intrinsic::ppc_altivec_lvxl:
    // Turn PPC lvx -> load if the pointer is known aligned.
    if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, DL) >= 16) {
      Value *Ptr = Builder->CreateBitCast(II->getArgOperand(0),
                                         PointerType::getUnqual(II->getType()));
      return new LoadInst(Ptr);
    }
    break;
  case Intrinsic::ppc_altivec_stvx:
  case Intrinsic::ppc_altivec_stvxl:
    // Turn stvx -> store if the pointer is known aligned.
    if (getOrEnforceKnownAlignment(II->getArgOperand(1), 16, DL) >= 16) {
      Type *OpPtrTy =
        PointerType::getUnqual(II->getArgOperand(0)->getType());
      Value *Ptr = Builder->CreateBitCast(II->getArgOperand(1), OpPtrTy);
      return new StoreInst(II->getArgOperand(0), Ptr);
    }
    break;
  case Intrinsic::x86_sse_storeu_ps:
  case Intrinsic::x86_sse2_storeu_pd:
  case Intrinsic::x86_sse2_storeu_dq:
    // Turn X86 storeu -> store if the pointer is known aligned.
    if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, DL) >= 16) {
      Type *OpPtrTy =
        PointerType::getUnqual(II->getArgOperand(1)->getType());
      Value *Ptr = Builder->CreateBitCast(II->getArgOperand(0), OpPtrTy);
      return new StoreInst(II->getArgOperand(1), Ptr);
    }
    break;

  case Intrinsic::x86_sse_cvtss2si:
  case Intrinsic::x86_sse_cvtss2si64:
  case Intrinsic::x86_sse_cvttss2si:
  case Intrinsic::x86_sse_cvttss2si64:
  case Intrinsic::x86_sse2_cvtsd2si:
  case Intrinsic::x86_sse2_cvtsd2si64:
  case Intrinsic::x86_sse2_cvttsd2si:
  case Intrinsic::x86_sse2_cvttsd2si64: {
    // These intrinsics only demand the 0th element of their input vectors. If
    // we can simplify the input based on that, do so now.
    unsigned VWidth =
      cast<VectorType>(II->getArgOperand(0)->getType())->getNumElements();
    APInt DemandedElts(VWidth, 1);
    APInt UndefElts(VWidth, 0);
    if (Value *V = SimplifyDemandedVectorElts(II->getArgOperand(0),
                                              DemandedElts, UndefElts)) {
      II->setArgOperand(0, V);
      return II;
    }
    break;
  }

  // Constant fold <A x Bi> << Ci.
  // FIXME: We don't handle _dq because it's a shift of an i128, but is
  // represented in the IR as <2 x i64>. A per element shift is wrong.
  case Intrinsic::x86_sse2_psll_d:
  case Intrinsic::x86_sse2_psll_q:
  case Intrinsic::x86_sse2_psll_w:
  case Intrinsic::x86_sse2_pslli_d:
  case Intrinsic::x86_sse2_pslli_q:
  case Intrinsic::x86_sse2_pslli_w:
  case Intrinsic::x86_avx2_psll_d:
  case Intrinsic::x86_avx2_psll_q:
  case Intrinsic::x86_avx2_psll_w:
  case Intrinsic::x86_avx2_pslli_d:
  case Intrinsic::x86_avx2_pslli_q:
  case Intrinsic::x86_avx2_pslli_w:
  case Intrinsic::x86_sse2_psrl_d:
  case Intrinsic::x86_sse2_psrl_q:
  case Intrinsic::x86_sse2_psrl_w:
  case Intrinsic::x86_sse2_psrli_d:
  case Intrinsic::x86_sse2_psrli_q:
  case Intrinsic::x86_sse2_psrli_w:
  case Intrinsic::x86_avx2_psrl_d:
  case Intrinsic::x86_avx2_psrl_q:
  case Intrinsic::x86_avx2_psrl_w:
  case Intrinsic::x86_avx2_psrli_d:
  case Intrinsic::x86_avx2_psrli_q:
  case Intrinsic::x86_avx2_psrli_w: {
    // Simplify if count is constant. To 0 if >= BitWidth,
    // otherwise to shl/lshr.
    auto CDV = dyn_cast<ConstantDataVector>(II->getArgOperand(1));
    auto CInt = dyn_cast<ConstantInt>(II->getArgOperand(1));
    if (!CDV && !CInt)
      break;
    ConstantInt *Count;
    if (CDV)
      Count = cast<ConstantInt>(CDV->getElementAsConstant(0));
    else
      Count = CInt;

    auto Vec = II->getArgOperand(0);
    auto VT = cast<VectorType>(Vec->getType());
    if (Count->getZExtValue() >
        VT->getElementType()->getPrimitiveSizeInBits() - 1)
      return ReplaceInstUsesWith(
          CI, ConstantAggregateZero::get(Vec->getType()));

    bool isPackedShiftLeft = true;
    switch (II->getIntrinsicID()) {
    default : break;
    case Intrinsic::x86_sse2_psrl_d:
    case Intrinsic::x86_sse2_psrl_q:
    case Intrinsic::x86_sse2_psrl_w:
    case Intrinsic::x86_sse2_psrli_d:
    case Intrinsic::x86_sse2_psrli_q:
    case Intrinsic::x86_sse2_psrli_w:
    case Intrinsic::x86_avx2_psrl_d:
    case Intrinsic::x86_avx2_psrl_q:
    case Intrinsic::x86_avx2_psrl_w:
    case Intrinsic::x86_avx2_psrli_d:
    case Intrinsic::x86_avx2_psrli_q:
    case Intrinsic::x86_avx2_psrli_w: isPackedShiftLeft = false; break;
    }

    unsigned VWidth = VT->getNumElements();
    // Get a constant vector of the same type as the first operand.
    auto VTCI = ConstantInt::get(VT->getElementType(), Count->getZExtValue());
    if (isPackedShiftLeft)
      return BinaryOperator::CreateShl(Vec,
          Builder->CreateVectorSplat(VWidth, VTCI));

    return BinaryOperator::CreateLShr(Vec,
        Builder->CreateVectorSplat(VWidth, VTCI));
  }

  case Intrinsic::x86_sse41_pmovsxbw:
  case Intrinsic::x86_sse41_pmovsxwd:
  case Intrinsic::x86_sse41_pmovsxdq:
  case Intrinsic::x86_sse41_pmovzxbw:
  case Intrinsic::x86_sse41_pmovzxwd:
  case Intrinsic::x86_sse41_pmovzxdq: {
    // pmov{s|z}x ignores the upper half of their input vectors.
    unsigned VWidth =
      cast<VectorType>(II->getArgOperand(0)->getType())->getNumElements();
    unsigned LowHalfElts = VWidth / 2;
    APInt InputDemandedElts(APInt::getBitsSet(VWidth, 0, LowHalfElts));
    APInt UndefElts(VWidth, 0);
    if (Value *TmpV = SimplifyDemandedVectorElts(II->getArgOperand(0),
                                                 InputDemandedElts,
                                                 UndefElts)) {
      II->setArgOperand(0, TmpV);
      return II;
    }
    break;
  }

  case Intrinsic::x86_sse4a_insertqi: {
    // insertqi x, y, 64, 0 can just copy y's lower bits and leave the top
    // ones undef
    // TODO: eventually we should lower this intrinsic to IR
    if (auto CIWidth = dyn_cast<ConstantInt>(II->getArgOperand(2))) {
      if (auto CIStart = dyn_cast<ConstantInt>(II->getArgOperand(3))) {
        if (CIWidth->equalsInt(64) && CIStart->isZero()) {
          Value *Vec = II->getArgOperand(1);
          Value *Undef = UndefValue::get(Vec->getType());
          const uint32_t Mask[] = { 0, 2 };
          return ReplaceInstUsesWith(
              CI,
              Builder->CreateShuffleVector(
                  Vec, Undef, ConstantDataVector::get(
                                  II->getContext(), ArrayRef<uint32_t>(Mask))));

        } else if (auto Source =
                       dyn_cast<IntrinsicInst>(II->getArgOperand(0))) {
          if (Source->hasOneUse() &&
              Source->getArgOperand(1) == II->getArgOperand(1)) {
            // If the source of the insert has only one use and it's another
            // insert (and they're both inserting from the same vector), try to
            // bundle both together.
            auto CISourceWidth =
                dyn_cast<ConstantInt>(Source->getArgOperand(2));
            auto CISourceStart =
                dyn_cast<ConstantInt>(Source->getArgOperand(3));
            if (CISourceStart && CISourceWidth) {
              unsigned Start = CIStart->getZExtValue();
              unsigned Width = CIWidth->getZExtValue();
              unsigned End = Start + Width;
              unsigned SourceStart = CISourceStart->getZExtValue();
              unsigned SourceWidth = CISourceWidth->getZExtValue();
              unsigned SourceEnd = SourceStart + SourceWidth;
              unsigned NewStart, NewWidth;
              bool ShouldReplace = false;
              if (Start <= SourceStart && SourceStart <= End) {
                NewStart = Start;
                NewWidth = std::max(End, SourceEnd) - NewStart;
                ShouldReplace = true;
              } else if (SourceStart <= Start && Start <= SourceEnd) {
                NewStart = SourceStart;
                NewWidth = std::max(SourceEnd, End) - NewStart;
                ShouldReplace = true;
              }

              if (ShouldReplace) {
                Constant *ConstantWidth = ConstantInt::get(
                    II->getArgOperand(2)->getType(), NewWidth, false);
                Constant *ConstantStart = ConstantInt::get(
                    II->getArgOperand(3)->getType(), NewStart, false);
                Value *Args[4] = { Source->getArgOperand(0),
                                   II->getArgOperand(1), ConstantWidth,
                                   ConstantStart };
                Module *M = CI.getParent()->getParent()->getParent();
                Value *F =
                    Intrinsic::getDeclaration(M, Intrinsic::x86_sse4a_insertqi);
                return ReplaceInstUsesWith(CI, Builder->CreateCall(F, Args));
              }
            }
          }
        }
      }
    }
    break;
  }

  case Intrinsic::x86_sse41_pblendvb:
  case Intrinsic::x86_sse41_blendvps:
  case Intrinsic::x86_sse41_blendvpd:
  case Intrinsic::x86_avx_blendv_ps_256:
  case Intrinsic::x86_avx_blendv_pd_256:
  case Intrinsic::x86_avx2_pblendvb: {
    // Convert blendv* to vector selects if the mask is constant.
    // This optimization is convoluted because the intrinsic is defined as
    // getting a vector of floats or doubles for the ps and pd versions.
    // FIXME: That should be changed.
    Value *Mask = II->getArgOperand(2);
    if (auto C = dyn_cast<ConstantDataVector>(Mask)) {
      auto Tyi1 = Builder->getInt1Ty();
      auto SelectorType = cast<VectorType>(Mask->getType());
      auto EltTy = SelectorType->getElementType();
      unsigned Size = SelectorType->getNumElements();
      unsigned BitWidth =
          EltTy->isFloatTy()
              ? 32
              : (EltTy->isDoubleTy() ? 64 : EltTy->getIntegerBitWidth());
      assert((BitWidth == 64 || BitWidth == 32 || BitWidth == 8) &&
             "Wrong arguments for variable blend intrinsic");
      SmallVector<Constant *, 32> Selectors;
      for (unsigned I = 0; I < Size; ++I) {
        // The intrinsics only read the top bit
        uint64_t Selector;
        if (BitWidth == 8)
          Selector = C->getElementAsInteger(I);
        else
          Selector = C->getElementAsAPFloat(I).bitcastToAPInt().getZExtValue();
        Selectors.push_back(ConstantInt::get(Tyi1, Selector >> (BitWidth - 1)));
      }
      auto NewSelector = ConstantVector::get(Selectors);
      return SelectInst::Create(NewSelector, II->getArgOperand(1),
                                II->getArgOperand(0), "blendv");
    } else {
      break;
    }
  }

  case Intrinsic::x86_avx_vpermilvar_ps:
  case Intrinsic::x86_avx_vpermilvar_ps_256:
  case Intrinsic::x86_avx_vpermilvar_pd:
  case Intrinsic::x86_avx_vpermilvar_pd_256: {
    // Convert vpermil* to shufflevector if the mask is constant.
    Value *V = II->getArgOperand(1);
    unsigned Size = cast<VectorType>(V->getType())->getNumElements();
    assert(Size == 8 || Size == 4 || Size == 2);
    uint32_t Indexes[8];
    if (auto C = dyn_cast<ConstantDataVector>(V)) {
      // The intrinsics only read one or two bits, clear the rest.
      for (unsigned I = 0; I < Size; ++I) {
        uint32_t Index = C->getElementAsInteger(I) & 0x3;
        if (II->getIntrinsicID() == Intrinsic::x86_avx_vpermilvar_pd ||
            II->getIntrinsicID() == Intrinsic::x86_avx_vpermilvar_pd_256)
          Index >>= 1;
        Indexes[I] = Index;
      }
    } else if (isa<ConstantAggregateZero>(V)) {
      for (unsigned I = 0; I < Size; ++I)
        Indexes[I] = 0;
    } else {
      break;
    }
    // The _256 variants are a bit trickier since the mask bits always index
    // into the corresponding 128 half. In order to convert to a generic
    // shuffle, we have to make that explicit.
    if (II->getIntrinsicID() == Intrinsic::x86_avx_vpermilvar_ps_256 ||
        II->getIntrinsicID() == Intrinsic::x86_avx_vpermilvar_pd_256) {
      for (unsigned I = Size / 2; I < Size; ++I)
        Indexes[I] += Size / 2;
    }
    auto NewC =
        ConstantDataVector::get(V->getContext(), makeArrayRef(Indexes, Size));
    auto V1 = II->getArgOperand(0);
    auto V2 = UndefValue::get(V1->getType());
    auto Shuffle = Builder->CreateShuffleVector(V1, V2, NewC);
    return ReplaceInstUsesWith(CI, Shuffle);
  }

  case Intrinsic::ppc_altivec_vperm:
    // Turn vperm(V1,V2,mask) -> shuffle(V1,V2,mask) if mask is a constant.
    // Note that ppc_altivec_vperm has a big-endian bias, so when creating
    // a vectorshuffle for little endian, we must undo the transformation
    // performed on vec_perm in altivec.h.  That is, we must complement
    // the permutation mask with respect to 31 and reverse the order of
    // V1 and V2.
    if (Constant *Mask = dyn_cast<Constant>(II->getArgOperand(2))) {
      assert(Mask->getType()->getVectorNumElements() == 16 &&
             "Bad type for intrinsic!");

      // Check that all of the elements are integer constants or undefs.
      bool AllEltsOk = true;
      for (unsigned i = 0; i != 16; ++i) {
        Constant *Elt = Mask->getAggregateElement(i);
        if (!Elt || !(isa<ConstantInt>(Elt) || isa<UndefValue>(Elt))) {
          AllEltsOk = false;
          break;
        }
      }

      if (AllEltsOk) {
        // Cast the input vectors to byte vectors.
        Value *Op0 = Builder->CreateBitCast(II->getArgOperand(0),
                                            Mask->getType());
        Value *Op1 = Builder->CreateBitCast(II->getArgOperand(1),
                                            Mask->getType());
        Value *Result = UndefValue::get(Op0->getType());

        // Only extract each element once.
        Value *ExtractedElts[32];
        memset(ExtractedElts, 0, sizeof(ExtractedElts));

        for (unsigned i = 0; i != 16; ++i) {
          if (isa<UndefValue>(Mask->getAggregateElement(i)))
            continue;
          unsigned Idx =
            cast<ConstantInt>(Mask->getAggregateElement(i))->getZExtValue();
          Idx &= 31;  // Match the hardware behavior.
          if (DL && DL->isLittleEndian())
            Idx = 31 - Idx;

          if (!ExtractedElts[Idx]) {
            Value *Op0ToUse = (DL && DL->isLittleEndian()) ? Op1 : Op0;
            Value *Op1ToUse = (DL && DL->isLittleEndian()) ? Op0 : Op1;
            ExtractedElts[Idx] =
              Builder->CreateExtractElement(Idx < 16 ? Op0ToUse : Op1ToUse,
                                            Builder->getInt32(Idx&15));
          }

          // Insert this value into the result vector.
          Result = Builder->CreateInsertElement(Result, ExtractedElts[Idx],
                                                Builder->getInt32(i));
        }
        return CastInst::Create(Instruction::BitCast, Result, CI.getType());
      }
    }
    break;

  case Intrinsic::arm_neon_vld1:
  case Intrinsic::arm_neon_vld2:
  case Intrinsic::arm_neon_vld3:
  case Intrinsic::arm_neon_vld4:
  case Intrinsic::arm_neon_vld2lane:
  case Intrinsic::arm_neon_vld3lane:
  case Intrinsic::arm_neon_vld4lane:
  case Intrinsic::arm_neon_vst1:
  case Intrinsic::arm_neon_vst2:
  case Intrinsic::arm_neon_vst3:
  case Intrinsic::arm_neon_vst4:
  case Intrinsic::arm_neon_vst2lane:
  case Intrinsic::arm_neon_vst3lane:
  case Intrinsic::arm_neon_vst4lane: {
    unsigned MemAlign = getKnownAlignment(II->getArgOperand(0), DL);
    unsigned AlignArg = II->getNumArgOperands() - 1;
    ConstantInt *IntrAlign = dyn_cast<ConstantInt>(II->getArgOperand(AlignArg));
    if (IntrAlign && IntrAlign->getZExtValue() < MemAlign) {
      II->setArgOperand(AlignArg,
                        ConstantInt::get(Type::getInt32Ty(II->getContext()),
                                         MemAlign, false));
      return II;
    }
    break;
  }

  case Intrinsic::arm_neon_vmulls:
  case Intrinsic::arm_neon_vmullu:
  case Intrinsic::aarch64_neon_smull:
  case Intrinsic::aarch64_neon_umull: {
    Value *Arg0 = II->getArgOperand(0);
    Value *Arg1 = II->getArgOperand(1);

    // Handle mul by zero first:
    if (isa<ConstantAggregateZero>(Arg0) || isa<ConstantAggregateZero>(Arg1)) {
      return ReplaceInstUsesWith(CI, ConstantAggregateZero::get(II->getType()));
    }

    // Check for constant LHS & RHS - in this case we just simplify.
    bool Zext = (II->getIntrinsicID() == Intrinsic::arm_neon_vmullu ||
                 II->getIntrinsicID() == Intrinsic::aarch64_neon_umull);
    VectorType *NewVT = cast<VectorType>(II->getType());
    if (Constant *CV0 = dyn_cast<Constant>(Arg0)) {
      if (Constant *CV1 = dyn_cast<Constant>(Arg1)) {
        CV0 = ConstantExpr::getIntegerCast(CV0, NewVT, /*isSigned=*/!Zext);
        CV1 = ConstantExpr::getIntegerCast(CV1, NewVT, /*isSigned=*/!Zext);

        return ReplaceInstUsesWith(CI, ConstantExpr::getMul(CV0, CV1));
      }

      // Couldn't simplify - canonicalize constant to the RHS.
      std::swap(Arg0, Arg1);
    }

    // Handle mul by one:
    if (Constant *CV1 = dyn_cast<Constant>(Arg1))
      if (ConstantInt *Splat =
              dyn_cast_or_null<ConstantInt>(CV1->getSplatValue()))
        if (Splat->isOne())
          return CastInst::CreateIntegerCast(Arg0, II->getType(),
                                             /*isSigned=*/!Zext);

    break;
  }

  case Intrinsic::AMDGPU_rcp: {
    if (const ConstantFP *C = dyn_cast<ConstantFP>(II->getArgOperand(0))) {
      const APFloat &ArgVal = C->getValueAPF();
      APFloat Val(ArgVal.getSemantics(), 1.0);
      APFloat::opStatus Status = Val.divide(ArgVal,
                                            APFloat::rmNearestTiesToEven);
      // Only do this if it was exact and therefore not dependent on the
      // rounding mode.
      if (Status == APFloat::opOK)
        return ReplaceInstUsesWith(CI, ConstantFP::get(II->getContext(), Val));
    }

    break;
  }
  case Intrinsic::stackrestore: {
    // If the save is right next to the restore, remove the restore.  This can
    // happen when variable allocas are DCE'd.
    if (IntrinsicInst *SS = dyn_cast<IntrinsicInst>(II->getArgOperand(0))) {
      if (SS->getIntrinsicID() == Intrinsic::stacksave) {
        BasicBlock::iterator BI = SS;
        if (&*++BI == II)
          return EraseInstFromFunction(CI);
      }
    }

    // Scan down this block to see if there is another stack restore in the
    // same block without an intervening call/alloca.
    BasicBlock::iterator BI = II;
    TerminatorInst *TI = II->getParent()->getTerminator();
    bool CannotRemove = false;
    for (++BI; &*BI != TI; ++BI) {
      if (isa<AllocaInst>(BI)) {
        CannotRemove = true;
        break;
      }
      if (CallInst *BCI = dyn_cast<CallInst>(BI)) {
        if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(BCI)) {
          // If there is a stackrestore below this one, remove this one.
          if (II->getIntrinsicID() == Intrinsic::stackrestore)
            return EraseInstFromFunction(CI);
          // Otherwise, ignore the intrinsic.
        } else {
          // If we found a non-intrinsic call, we can't remove the stack
          // restore.
          CannotRemove = true;
          break;
        }
      }
    }

    // If the stack restore is in a return, resume, or unwind block and if there
    // are no allocas or calls between the restore and the return, nuke the
    // restore.
    if (!CannotRemove && (isa<ReturnInst>(TI) || isa<ResumeInst>(TI)))
      return EraseInstFromFunction(CI);
    break;
  }
  }

  return visitCallSite(II);
}

Instruction * InstCombiner::visitCallSite ( CallSite  CS )

private

Definition at line 1126 of file InstCombineCalls.cpp.

                                                    {
  if (isAllocLikeFn(CS.getInstruction(), TLI))
    return visitAllocSite(*CS.getInstruction());

  bool Changed = false;

  // If the callee is a pointer to a function, attempt to move any casts to the
  // arguments of the call/invoke.
  Value *Callee = CS.getCalledValue();
  if (!isa<Function>(Callee) && transformConstExprCastCall(CS))
    return nullptr;

  if (Function *CalleeF = dyn_cast<Function>(Callee))
    // If the call and callee calling conventions don't match, this call must
    // be unreachable, as the call is undefined.
    if (CalleeF->getCallingConv() != CS.getCallingConv() &&
        // Only do this for calls to a function with a body.  A prototype may
        // not actually end up matching the implementation's calling conv for a
        // variety of reasons (e.g. it may be written in assembly).
        !CalleeF->isDeclaration()) {
      Instruction *OldCall = CS.getInstruction();
      new StoreInst(ConstantInt::getTrue(Callee->getContext()),
                UndefValue::get(Type::getInt1PtrTy(Callee->getContext())),
                                  OldCall);
      // If OldCall does not return void then replaceAllUsesWith undef.
      // This allows ValueHandlers and custom metadata to adjust itself.
      if (!OldCall->getType()->isVoidTy())
        ReplaceInstUsesWith(*OldCall, UndefValue::get(OldCall->getType()));
      if (isa<CallInst>(OldCall))
        return EraseInstFromFunction(*OldCall);

      // We cannot remove an invoke, because it would change the CFG, just
      // change the callee to a null pointer.
      cast<InvokeInst>(OldCall)->setCalledFunction(
                                    Constant::getNullValue(CalleeF->getType()));
      return nullptr;
    }

  if (isa<ConstantPointerNull>(Callee) || isa<UndefValue>(Callee)) {
    // If CS does not return void then replaceAllUsesWith undef.
    // This allows ValueHandlers and custom metadata to adjust itself.
    if (!CS.getInstruction()->getType()->isVoidTy())
      ReplaceInstUsesWith(*CS.getInstruction(),
                          UndefValue::get(CS.getInstruction()->getType()));

    if (isa<InvokeInst>(CS.getInstruction())) {
      // Can't remove an invoke because we cannot change the CFG.
      return nullptr;
    }

    // This instruction is not reachable, just remove it.  We insert a store to
    // undef so that we know that this code is not reachable, despite the fact
    // that we can't modify the CFG here.
    new StoreInst(ConstantInt::getTrue(Callee->getContext()),
                  UndefValue::get(Type::getInt1PtrTy(Callee->getContext())),
                  CS.getInstruction());

    return EraseInstFromFunction(*CS.getInstruction());
  }

  if (IntrinsicInst *II = FindInitTrampoline(Callee))
    return transformCallThroughTrampoline(CS, II);

  PointerType *PTy = cast<PointerType>(Callee->getType());
  FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
  if (FTy->isVarArg()) {
    int ix = FTy->getNumParams();
    // See if we can optimize any arguments passed through the varargs area of
    // the call.
    for (CallSite::arg_iterator I = CS.arg_begin() + FTy->getNumParams(),
           E = CS.arg_end(); I != E; ++I, ++ix) {
      CastInst *CI = dyn_cast<CastInst>(*I);
      if (CI && isSafeToEliminateVarargsCast(CS, CI, DL, ix)) {
        *I = CI->getOperand(0);
        Changed = true;
      }
    }
  }

  if (isa<InlineAsm>(Callee) && !CS.doesNotThrow()) {
    // Inline asm calls cannot throw - mark them 'nounwind'.
    CS.setDoesNotThrow();
    Changed = true;
  }

  // Try to optimize the call if possible, we require DataLayout for most of
  // this.  None of these calls are seen as possibly dead so go ahead and
  // delete the instruction now.
  if (CallInst *CI = dyn_cast<CallInst>(CS.getInstruction())) {
    Instruction *I = tryOptimizeCall(CI, DL);
    // If we changed something return the result, etc. Otherwise let
    // the fallthrough check.
    if (I) return EraseInstFromFunction(*I);
  }

  return Changed ? CS.getInstruction() : nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitCastInst ( CastInst &  I )

inlineinherited

Definition at line 231 of file InstVisitor.h.

{ DELEGATE(UnaryInstruction);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitCmpInst ( CmpInst &  I )

inlineinherited

Definition at line 233 of file InstVisitor.h.

{ DELEGATE(Instruction);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitDbgDeclareInst ( DbgDeclareInst &  I )

inlineinherited

Definition at line 205 of file InstVisitor.h.

{ DELEGATE(DbgInfoIntrinsic);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitDbgInfoIntrinsic ( DbgInfoIntrinsic &  I )

inlineinherited

Definition at line 207 of file InstVisitor.h.

{ DELEGATE(IntrinsicInst); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitDbgValueInst ( DbgValueInst &  I )

inlineinherited

Definition at line 206 of file InstVisitor.h.

{ DELEGATE(DbgInfoIntrinsic);}

Instruction * InstCombiner::visitExtractElementInst

(

ExtractElementInst &

EI

)

Definition at line 179 of file InstCombineVectorOps.cpp.

                                                                         {
  // If vector val is constant with all elements the same, replace EI with
  // that element.  We handle a known element # below.
  if (Constant *C = dyn_cast<Constant>(EI.getOperand(0)))
    if (CheapToScalarize(C, false))
      return ReplaceInstUsesWith(EI, C->getAggregateElement(0U));

  // If extracting a specified index from the vector, see if we can recursively
  // find a previously computed scalar that was inserted into the vector.
  if (ConstantInt *IdxC = dyn_cast<ConstantInt>(EI.getOperand(1))) {
    unsigned IndexVal = IdxC->getZExtValue();
    unsigned VectorWidth = EI.getVectorOperandType()->getNumElements();

    // If this is extracting an invalid index, turn this into undef, to avoid
    // crashing the code below.
    if (IndexVal >= VectorWidth)
      return ReplaceInstUsesWith(EI, UndefValue::get(EI.getType()));

    // This instruction only demands the single element from the input vector.
    // If the input vector has a single use, simplify it based on this use
    // property.
    if (EI.getOperand(0)->hasOneUse() && VectorWidth != 1) {
      APInt UndefElts(VectorWidth, 0);
      APInt DemandedMask(VectorWidth, 0);
      DemandedMask.setBit(IndexVal);
      if (Value *V = SimplifyDemandedVectorElts(EI.getOperand(0),
                                                DemandedMask, UndefElts)) {
        EI.setOperand(0, V);
        return &EI;
      }
    }

    if (Value *Elt = FindScalarElement(EI.getOperand(0), IndexVal))
      return ReplaceInstUsesWith(EI, Elt);

    // If the this extractelement is directly using a bitcast from a vector of
    // the same number of elements, see if we can find the source element from
    // it.  In this case, we will end up needing to bitcast the scalars.
    if (BitCastInst *BCI = dyn_cast<BitCastInst>(EI.getOperand(0))) {
      if (VectorType *VT = dyn_cast<VectorType>(BCI->getOperand(0)->getType()))
        if (VT->getNumElements() == VectorWidth)
          if (Value *Elt = FindScalarElement(BCI->getOperand(0), IndexVal))
            return new BitCastInst(Elt, EI.getType());
    }

    // If there's a vector PHI feeding a scalar use through this extractelement
    // instruction, try to scalarize the PHI.
    if (PHINode *PN = dyn_cast<PHINode>(EI.getOperand(0))) {
      Instruction *scalarPHI = scalarizePHI(EI, PN);
      if (scalarPHI)
        return scalarPHI;
    }
  }

  if (Instruction *I = dyn_cast<Instruction>(EI.getOperand(0))) {
    // Push extractelement into predecessor operation if legal and
    // profitable to do so
    if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
      if (I->hasOneUse() &&
          CheapToScalarize(BO, isa<ConstantInt>(EI.getOperand(1)))) {
        Value *newEI0 =
          Builder->CreateExtractElement(BO->getOperand(0), EI.getOperand(1),
                                        EI.getName()+".lhs");
        Value *newEI1 =
          Builder->CreateExtractElement(BO->getOperand(1), EI.getOperand(1),
                                        EI.getName()+".rhs");
        return BinaryOperator::Create(BO->getOpcode(), newEI0, newEI1);
      }
    } else if (InsertElementInst *IE = dyn_cast<InsertElementInst>(I)) {
      // Extracting the inserted element?
      if (IE->getOperand(2) == EI.getOperand(1))
        return ReplaceInstUsesWith(EI, IE->getOperand(1));
      // If the inserted and extracted elements are constants, they must not
      // be the same value, extract from the pre-inserted value instead.
      if (isa<Constant>(IE->getOperand(2)) && isa<Constant>(EI.getOperand(1))) {
        Worklist.AddValue(EI.getOperand(0));
        EI.setOperand(0, IE->getOperand(0));
        return &EI;
      }
    } else if (ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(I)) {
      // If this is extracting an element from a shufflevector, figure out where
      // it came from and extract from the appropriate input element instead.
      if (ConstantInt *Elt = dyn_cast<ConstantInt>(EI.getOperand(1))) {
        int SrcIdx = SVI->getMaskValue(Elt->getZExtValue());
        Value *Src;
        unsigned LHSWidth =
          SVI->getOperand(0)->getType()->getVectorNumElements();

        if (SrcIdx < 0)
          return ReplaceInstUsesWith(EI, UndefValue::get(EI.getType()));
        if (SrcIdx < (int)LHSWidth)
          Src = SVI->getOperand(0);
        else {
          SrcIdx -= LHSWidth;
          Src = SVI->getOperand(1);
        }
        Type *Int32Ty = Type::getInt32Ty(EI.getContext());
        return ExtractElementInst::Create(Src,
                                          ConstantInt::get(Int32Ty,
                                                           SrcIdx, false));
      }
    } else if (CastInst *CI = dyn_cast<CastInst>(I)) {
      // Canonicalize extractelement(cast) -> cast(extractelement)
      // bitcasts can change the number of vector elements and they cost nothing
      if (CI->hasOneUse() && (CI->getOpcode() != Instruction::BitCast)) {
        Value *EE = Builder->CreateExtractElement(CI->getOperand(0),
                                                  EI.getIndexOperand());
        Worklist.AddValue(EE);
        return CastInst::Create(CI->getOpcode(), EE, EI.getType());
      }
    } else if (SelectInst *SI = dyn_cast<SelectInst>(I)) {
      if (SI->hasOneUse()) {
        // TODO: For a select on vectors, it might be useful to do this if it
        // has multiple extractelement uses. For vector select, that seems to
        // fight the vectorizer.

        // If we are extracting an element from a vector select or a select on
        // vectors, a select on the scalars extracted from the vector arguments.
        Value *TrueVal = SI->getTrueValue();
        Value *FalseVal = SI->getFalseValue();

        Value *Cond = SI->getCondition();
        if (Cond->getType()->isVectorTy()) {
          Cond = Builder->CreateExtractElement(Cond,
                                               EI.getIndexOperand(),
                                               Cond->getName() + ".elt");
        }

        Value *V1Elem
          = Builder->CreateExtractElement(TrueVal,
                                          EI.getIndexOperand(),
                                          TrueVal->getName() + ".elt");

        Value *V2Elem
          = Builder->CreateExtractElement(FalseVal,
                                          EI.getIndexOperand(),
                                          FalseVal->getName() + ".elt");
        return SelectInst::Create(Cond,
                                  V1Elem,
                                  V2Elem,
                                  SI->getName() + ".elt");
      }
    }
  }
  return nullptr;
}

Instruction * InstCombiner::visitExtractValueInst

(

ExtractValueInst &

EV

)

Definition at line 1999 of file InstructionCombining.cpp.

                                                                     {
  Value *Agg = EV.getAggregateOperand();

  if (!EV.hasIndices())
    return ReplaceInstUsesWith(EV, Agg);

  if (Constant *C = dyn_cast<Constant>(Agg)) {
    if (Constant *C2 = C->getAggregateElement(*EV.idx_begin())) {
      if (EV.getNumIndices() == 0)
        return ReplaceInstUsesWith(EV, C2);
      // Extract the remaining indices out of the constant indexed by the
      // first index
      return ExtractValueInst::Create(C2, EV.getIndices().slice(1));
    }
    return nullptr; // Can't handle other constants
  }

  if (InsertValueInst *IV = dyn_cast<InsertValueInst>(Agg)) {
    // We're extracting from an insertvalue instruction, compare the indices
    const unsigned *exti, *exte, *insi, *inse;
    for (exti = EV.idx_begin(), insi = IV->idx_begin(),
         exte = EV.idx_end(), inse = IV->idx_end();
         exti != exte && insi != inse;
         ++exti, ++insi) {
      if (*insi != *exti)
        // The insert and extract both reference distinctly different elements.
        // This means the extract is not influenced by the insert, and we can
        // replace the aggregate operand of the extract with the aggregate
        // operand of the insert. i.e., replace
        // %I = insertvalue { i32, { i32 } } %A, { i32 } { i32 42 }, 1
        // %E = extractvalue { i32, { i32 } } %I, 0
        // with
        // %E = extractvalue { i32, { i32 } } %A, 0
        return ExtractValueInst::Create(IV->getAggregateOperand(),
                                        EV.getIndices());
    }
    if (exti == exte && insi == inse)
      // Both iterators are at the end: Index lists are identical. Replace
      // %B = insertvalue { i32, { i32 } } %A, i32 42, 1, 0
      // %C = extractvalue { i32, { i32 } } %B, 1, 0
      // with "i32 42"
      return ReplaceInstUsesWith(EV, IV->getInsertedValueOperand());
    if (exti == exte) {
      // The extract list is a prefix of the insert list. i.e. replace
      // %I = insertvalue { i32, { i32 } } %A, i32 42, 1, 0
      // %E = extractvalue { i32, { i32 } } %I, 1
      // with
      // %X = extractvalue { i32, { i32 } } %A, 1
      // %E = insertvalue { i32 } %X, i32 42, 0
      // by switching the order of the insert and extract (though the
      // insertvalue should be left in, since it may have other uses).
      Value *NewEV = Builder->CreateExtractValue(IV->getAggregateOperand(),
                                                 EV.getIndices());
      return InsertValueInst::Create(NewEV, IV->getInsertedValueOperand(),
                                     makeArrayRef(insi, inse));
    }
    if (insi == inse)
      // The insert list is a prefix of the extract list
      // We can simply remove the common indices from the extract and make it
      // operate on the inserted value instead of the insertvalue result.
      // i.e., replace
      // %I = insertvalue { i32, { i32 } } %A, { i32 } { i32 42 }, 1
      // %E = extractvalue { i32, { i32 } } %I, 1, 0
      // with
      // %E extractvalue { i32 } { i32 42 }, 0
      return ExtractValueInst::Create(IV->getInsertedValueOperand(),
                                      makeArrayRef(exti, exte));
  }
  if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Agg)) {
    // We're extracting from an intrinsic, see if we're the only user, which
    // allows us to simplify multiple result intrinsics to simpler things that
    // just get one value.
    if (II->hasOneUse()) {
      // Check if we're grabbing the overflow bit or the result of a 'with
      // overflow' intrinsic.  If it's the latter we can remove the intrinsic
      // and replace it with a traditional binary instruction.
      switch (II->getIntrinsicID()) {
      case Intrinsic::uadd_with_overflow:
      case Intrinsic::sadd_with_overflow:
        if (*EV.idx_begin() == 0) {  // Normal result.
          Value *LHS = II->getArgOperand(0), *RHS = II->getArgOperand(1);
          ReplaceInstUsesWith(*II, UndefValue::get(II->getType()));
          EraseInstFromFunction(*II);
          return BinaryOperator::CreateAdd(LHS, RHS);
        }

        // If the normal result of the add is dead, and the RHS is a constant,
        // we can transform this into a range comparison.
        // overflow = uadd a, -4  -->  overflow = icmp ugt a, 3
        if (II->getIntrinsicID() == Intrinsic::uadd_with_overflow)
          if (ConstantInt *CI = dyn_cast<ConstantInt>(II->getArgOperand(1)))
            return new ICmpInst(ICmpInst::ICMP_UGT, II->getArgOperand(0),
                                ConstantExpr::getNot(CI));
        break;
      case Intrinsic::usub_with_overflow:
      case Intrinsic::ssub_with_overflow:
        if (*EV.idx_begin() == 0) {  // Normal result.
          Value *LHS = II->getArgOperand(0), *RHS = II->getArgOperand(1);
          ReplaceInstUsesWith(*II, UndefValue::get(II->getType()));
          EraseInstFromFunction(*II);
          return BinaryOperator::CreateSub(LHS, RHS);
        }
        break;
      case Intrinsic::umul_with_overflow:
      case Intrinsic::smul_with_overflow:
        if (*EV.idx_begin() == 0) {  // Normal result.
          Value *LHS = II->getArgOperand(0), *RHS = II->getArgOperand(1);
          ReplaceInstUsesWith(*II, UndefValue::get(II->getType()));
          EraseInstFromFunction(*II);
          return BinaryOperator::CreateMul(LHS, RHS);
        }
        break;
      default:
        break;
      }
    }
  }
  if (LoadInst *L = dyn_cast<LoadInst>(Agg))
    // If the (non-volatile) load only has one use, we can rewrite this to a
    // load from a GEP. This reduces the size of the load.
    // FIXME: If a load is used only by extractvalue instructions then this
    //        could be done regardless of having multiple uses.
    if (L->isSimple() && L->hasOneUse()) {
      // extractvalue has integer indices, getelementptr has Value*s. Convert.
      SmallVector<Value*, 4> Indices;
      // Prefix an i32 0 since we need the first element.
      Indices.push_back(Builder->getInt32(0));
      for (ExtractValueInst::idx_iterator I = EV.idx_begin(), E = EV.idx_end();
            I != E; ++I)
        Indices.push_back(Builder->getInt32(*I));

      // We need to insert these at the location of the old load, not at that of
      // the extractvalue.
      Builder->SetInsertPoint(L->getParent(), L);
      Value *GEP = Builder->CreateInBoundsGEP(L->getPointerOperand(), Indices);
      // Returning the load directly will cause the main loop to insert it in
      // the wrong spot, so use ReplaceInstUsesWith().
      return ReplaceInstUsesWith(EV, Builder->CreateLoad(GEP));
    }
  // We could simplify extracts from other values. Note that nested extracts may
  // already be simplified implicitly by the above: extract (extract (insert) )
  // will be translated into extract ( insert ( extract ) ) first and then just
  // the value inserted, if appropriate. Similarly for extracts from single-use
  // loads: extract (extract (load)) will be translated to extract (load (gep))
  // and if again single-use then via load (gep (gep)) to load (gep).
  // However, double extracts from e.g. function arguments or return values
  // aren't handled yet.
  return nullptr;
}

Instruction * InstCombiner::visitFAdd

(

BinaryOperator &

I

)

Definition at line 1272 of file InstCombineAddSub.cpp.

                                                      {
  bool Changed = SimplifyAssociativeOrCommutative(I);
  Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyFAddInst(LHS, RHS, I.getFastMathFlags(), DL))
    return ReplaceInstUsesWith(I, V);

  if (isa<Constant>(RHS)) {
    if (isa<PHINode>(LHS))
      if (Instruction *NV = FoldOpIntoPhi(I))
        return NV;

    if (SelectInst *SI = dyn_cast<SelectInst>(LHS))
      if (Instruction *NV = FoldOpIntoSelect(I, SI))
        return NV;
  }

  // -A + B  -->  B - A
  // -A + -B  -->  -(A + B)
  if (Value *LHSV = dyn_castFNegVal(LHS)) {
    Instruction *RI = BinaryOperator::CreateFSub(RHS, LHSV);
    RI->copyFastMathFlags(&I);
    return RI;
  }

  // A + -B  -->  A - B
  if (!isa<Constant>(RHS))
    if (Value *V = dyn_castFNegVal(RHS)) {
      Instruction *RI = BinaryOperator::CreateFSub(LHS, V);
      RI->copyFastMathFlags(&I);
      return RI;
    }

  // Check for (fadd double (sitofp x), y), see if we can merge this into an
  // integer add followed by a promotion.
  if (SIToFPInst *LHSConv = dyn_cast<SIToFPInst>(LHS)) {
    // (fadd double (sitofp x), fpcst) --> (sitofp (add int x, intcst))
    // ... if the constant fits in the integer value.  This is useful for things
    // like (double)(x & 1234) + 4.0 -> (double)((X & 1234)+4) which no longer
    // requires a constant pool load, and generally allows the add to be better
    // instcombined.
    if (ConstantFP *CFP = dyn_cast<ConstantFP>(RHS)) {
      Constant *CI =
      ConstantExpr::getFPToSI(CFP, LHSConv->getOperand(0)->getType());
      if (LHSConv->hasOneUse() &&
          ConstantExpr::getSIToFP(CI, I.getType()) == CFP &&
          WillNotOverflowSignedAdd(LHSConv->getOperand(0), CI)) {
        // Insert the new integer add.
        Value *NewAdd = Builder->CreateNSWAdd(LHSConv->getOperand(0),
                                              CI, "addconv");
        return new SIToFPInst(NewAdd, I.getType());
      }
    }

    // (fadd double (sitofp x), (sitofp y)) --> (sitofp (add int x, y))
    if (SIToFPInst *RHSConv = dyn_cast<SIToFPInst>(RHS)) {
      // Only do this if x/y have the same type, if at last one of them has a
      // single use (so we don't increase the number of int->fp conversions),
      // and if the integer add will not overflow.
      if (LHSConv->getOperand(0)->getType()==RHSConv->getOperand(0)->getType()&&
          (LHSConv->hasOneUse() || RHSConv->hasOneUse()) &&
          WillNotOverflowSignedAdd(LHSConv->getOperand(0),
                                   RHSConv->getOperand(0))) {
        // Insert the new integer add.
        Value *NewAdd = Builder->CreateNSWAdd(LHSConv->getOperand(0),
                                              RHSConv->getOperand(0),"addconv");
        return new SIToFPInst(NewAdd, I.getType());
      }
    }
  }

  // select C, 0, B + select C, A, 0 -> select C, A, B
  {
    Value *A1, *B1, *C1, *A2, *B2, *C2;
    if (match(LHS, m_Select(m_Value(C1), m_Value(A1), m_Value(B1))) &&
        match(RHS, m_Select(m_Value(C2), m_Value(A2), m_Value(B2)))) {
      if (C1 == C2) {
        Constant *Z1=nullptr, *Z2=nullptr;
        Value *A, *B, *C=C1;
        if (match(A1, m_AnyZero()) && match(B2, m_AnyZero())) {
            Z1 = dyn_cast<Constant>(A1); A = A2;
            Z2 = dyn_cast<Constant>(B2); B = B1;
        } else if (match(B1, m_AnyZero()) && match(A2, m_AnyZero())) {
            Z1 = dyn_cast<Constant>(B1); B = B2;
            Z2 = dyn_cast<Constant>(A2); A = A1; 
        }
        
        if (Z1 && Z2 && 
            (I.hasNoSignedZeros() || 
             (Z1->isNegativeZeroValue() && Z2->isNegativeZeroValue()))) {
          return SelectInst::Create(C, A, B);
        }
      }
    }
  }

  if (I.hasUnsafeAlgebra()) {
    if (Value *V = FAddCombine(Builder).simplify(&I))
      return ReplaceInstUsesWith(I, V);
  }

  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitFCmpInst

(

FCmpInst &

I

)

Orders the operands of the compare so that they are listed from most complex to least complex. This puts constants before unary operators, before binary operators.

Definition at line 3492 of file InstCombineCompares.cpp.

                                                    {
  bool Changed = false;

  /// Orders the operands of the compare so that they are listed from most
  /// complex to least complex.  This puts constants before unary operators,
  /// before binary operators.
  if (getComplexity(I.getOperand(0)) < getComplexity(I.getOperand(1))) {
    I.swapOperands();
    Changed = true;
  }

  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyFCmpInst(I.getPredicate(), Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // Simplify 'fcmp pred X, X'
  if (Op0 == Op1) {
    switch (I.getPredicate()) {
    default: llvm_unreachable("Unknown predicate!");
    case FCmpInst::FCMP_UNO:    // True if unordered: isnan(X) | isnan(Y)
    case FCmpInst::FCMP_ULT:    // True if unordered or less than
    case FCmpInst::FCMP_UGT:    // True if unordered or greater than
    case FCmpInst::FCMP_UNE:    // True if unordered or not equal
      // Canonicalize these to be 'fcmp uno %X, 0.0'.
      I.setPredicate(FCmpInst::FCMP_UNO);
      I.setOperand(1, Constant::getNullValue(Op0->getType()));
      return &I;

    case FCmpInst::FCMP_ORD:    // True if ordered (no nans)
    case FCmpInst::FCMP_OEQ:    // True if ordered and equal
    case FCmpInst::FCMP_OGE:    // True if ordered and greater than or equal
    case FCmpInst::FCMP_OLE:    // True if ordered and less than or equal
      // Canonicalize these to be 'fcmp ord %X, 0.0'.
      I.setPredicate(FCmpInst::FCMP_ORD);
      I.setOperand(1, Constant::getNullValue(Op0->getType()));
      return &I;
    }
  }

  // Handle fcmp with constant RHS
  if (Constant *RHSC = dyn_cast<Constant>(Op1)) {
    if (Instruction *LHSI = dyn_cast<Instruction>(Op0))
      switch (LHSI->getOpcode()) {
      case Instruction::FPExt: {
        // fcmp (fpext x), C -> fcmp x, (fptrunc C) if fptrunc is lossless
        FPExtInst *LHSExt = cast<FPExtInst>(LHSI);
        ConstantFP *RHSF = dyn_cast<ConstantFP>(RHSC);
        if (!RHSF)
          break;

        const fltSemantics *Sem;
        // FIXME: This shouldn't be here.
        if (LHSExt->getSrcTy()->isHalfTy())
          Sem = &APFloat::IEEEhalf;
        else if (LHSExt->getSrcTy()->isFloatTy())
          Sem = &APFloat::IEEEsingle;
        else if (LHSExt->getSrcTy()->isDoubleTy())
          Sem = &APFloat::IEEEdouble;
        else if (LHSExt->getSrcTy()->isFP128Ty())
          Sem = &APFloat::IEEEquad;
        else if (LHSExt->getSrcTy()->isX86_FP80Ty())
          Sem = &APFloat::x87DoubleExtended;
        else if (LHSExt->getSrcTy()->isPPC_FP128Ty())
          Sem = &APFloat::PPCDoubleDouble;
        else
          break;

        bool Lossy;
        APFloat F = RHSF->getValueAPF();
        F.convert(*Sem, APFloat::rmNearestTiesToEven, &Lossy);

        // Avoid lossy conversions and denormals. Zero is a special case
        // that's OK to convert.
        APFloat Fabs = F;
        Fabs.clearSign();
        if (!Lossy &&
            ((Fabs.compare(APFloat::getSmallestNormalized(*Sem)) !=
                 APFloat::cmpLessThan) || Fabs.isZero()))

          return new FCmpInst(I.getPredicate(), LHSExt->getOperand(0),
                              ConstantFP::get(RHSC->getContext(), F));
        break;
      }
      case Instruction::PHI:
        // Only fold fcmp into the PHI if the phi and fcmp are in the same
        // block.  If in the same block, we're encouraging jump threading.  If
        // not, we are just pessimizing the code by making an i1 phi.
        if (LHSI->getParent() == I.getParent())
          if (Instruction *NV = FoldOpIntoPhi(I))
            return NV;
        break;
      case Instruction::SIToFP:
      case Instruction::UIToFP:
        if (Instruction *NV = FoldFCmp_IntToFP_Cst(I, LHSI, RHSC))
          return NV;
        break;
      case Instruction::FSub: {
        // fcmp pred (fneg x), C -> fcmp swap(pred) x, -C
        Value *Op;
        if (match(LHSI, m_FNeg(m_Value(Op))))
          return new FCmpInst(I.getSwappedPredicate(), Op,
                              ConstantExpr::getFNeg(RHSC));
        break;
      }
      case Instruction::Load:
        if (GetElementPtrInst *GEP =
            dyn_cast<GetElementPtrInst>(LHSI->getOperand(0))) {
          if (GlobalVariable *GV = dyn_cast<GlobalVariable>(GEP->getOperand(0)))
            if (GV->isConstant() && GV->hasDefinitiveInitializer() &&
                !cast<LoadInst>(LHSI)->isVolatile())
              if (Instruction *Res = FoldCmpLoadFromIndexedGlobal(GEP, GV, I))
                return Res;
        }
        break;
      case Instruction::Call: {
        CallInst *CI = cast<CallInst>(LHSI);
        LibFunc::Func Func;
        // Various optimization for fabs compared with zero.
        if (RHSC->isNullValue() && CI->getCalledFunction() &&
            TLI->getLibFunc(CI->getCalledFunction()->getName(), Func) &&
            TLI->has(Func)) {
          if (Func == LibFunc::fabs || Func == LibFunc::fabsf ||
              Func == LibFunc::fabsl) {
            switch (I.getPredicate()) {
            default: break;
            // fabs(x) < 0 --> false
            case FCmpInst::FCMP_OLT:
              return ReplaceInstUsesWith(I, Builder->getFalse());
            // fabs(x) > 0 --> x != 0
            case FCmpInst::FCMP_OGT:
              return new FCmpInst(FCmpInst::FCMP_ONE, CI->getArgOperand(0),
                                  RHSC);
            // fabs(x) <= 0 --> x == 0
            case FCmpInst::FCMP_OLE:
              return new FCmpInst(FCmpInst::FCMP_OEQ, CI->getArgOperand(0),
                                  RHSC);
            // fabs(x) >= 0 --> !isnan(x)
            case FCmpInst::FCMP_OGE:
              return new FCmpInst(FCmpInst::FCMP_ORD, CI->getArgOperand(0),
                                  RHSC);
            // fabs(x) == 0 --> x == 0
            // fabs(x) != 0 --> x != 0
            case FCmpInst::FCMP_OEQ:
            case FCmpInst::FCMP_UEQ:
            case FCmpInst::FCMP_ONE:
            case FCmpInst::FCMP_UNE:
              return new FCmpInst(I.getPredicate(), CI->getArgOperand(0),
                                  RHSC);
            }
          }
        }
      }
      }
  }

  // fcmp pred (fneg x), (fneg y) -> fcmp swap(pred) x, y
  Value *X, *Y;
  if (match(Op0, m_FNeg(m_Value(X))) && match(Op1, m_FNeg(m_Value(Y))))
    return new FCmpInst(I.getSwappedPredicate(), X, Y);

  // fcmp (fpext x), (fpext y) -> fcmp x, y
  if (FPExtInst *LHSExt = dyn_cast<FPExtInst>(Op0))
    if (FPExtInst *RHSExt = dyn_cast<FPExtInst>(Op1))
      if (LHSExt->getSrcTy() == RHSExt->getSrcTy())
        return new FCmpInst(I.getPredicate(), LHSExt->getOperand(0),
                            RHSExt->getOperand(0));

  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitFDiv

(

BinaryOperator &

I

)

Definition at line 1061 of file InstCombineMulDivRem.cpp.

                                                      {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyFDivInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  if (isa<Constant>(Op0))
    if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;

  bool AllowReassociate = I.hasUnsafeAlgebra();
  bool AllowReciprocal = I.hasAllowReciprocal();

  if (Constant *Op1C = dyn_cast<Constant>(Op1)) {
    if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;

    if (AllowReassociate) {
      Constant *C1 = nullptr;
      Constant *C2 = Op1C;
      Value *X;
      Instruction *Res = nullptr;

      if (match(Op0, m_FMul(m_Value(X), m_Constant(C1)))) {
        // (X*C1)/C2 => X * (C1/C2)
        //
        Constant *C = ConstantExpr::getFDiv(C1, C2);
        if (isNormalFp(C))
          Res = BinaryOperator::CreateFMul(X, C);
      } else if (match(Op0, m_FDiv(m_Value(X), m_Constant(C1)))) {
        // (X/C1)/C2 => X /(C2*C1) [=> X * 1/(C2*C1) if reciprocal is allowed]
        //
        Constant *C = ConstantExpr::getFMul(C1, C2);
        if (isNormalFp(C)) {
          Res = CvtFDivConstToReciprocal(X, C, AllowReciprocal);
          if (!Res)
            Res = BinaryOperator::CreateFDiv(X, C);
        }
      }

      if (Res) {
        Res->setFastMathFlags(I.getFastMathFlags());
        return Res;
      }
    }

    // X / C => X * 1/C
    if (Instruction *T = CvtFDivConstToReciprocal(Op0, Op1C, AllowReciprocal)) {
      T->copyFastMathFlags(&I);
      return T;
    }

    return nullptr;
  }

  if (AllowReassociate && isa<Constant>(Op0)) {
    Constant *C1 = cast<Constant>(Op0), *C2;
    Constant *Fold = nullptr;
    Value *X;
    bool CreateDiv = true;

    // C1 / (X*C2) => (C1/C2) / X
    if (match(Op1, m_FMul(m_Value(X), m_Constant(C2))))
      Fold = ConstantExpr::getFDiv(C1, C2);
    else if (match(Op1, m_FDiv(m_Value(X), m_Constant(C2)))) {
      // C1 / (X/C2) => (C1*C2) / X
      Fold = ConstantExpr::getFMul(C1, C2);
    } else if (match(Op1, m_FDiv(m_Constant(C2), m_Value(X)))) {
      // C1 / (C2/X) => (C1/C2) * X
      Fold = ConstantExpr::getFDiv(C1, C2);
      CreateDiv = false;
    }

    if (Fold && isNormalFp(Fold)) {
      Instruction *R = CreateDiv ? BinaryOperator::CreateFDiv(Fold, X)
                                 : BinaryOperator::CreateFMul(X, Fold);
      R->setFastMathFlags(I.getFastMathFlags());
      return R;
    }
    return nullptr;
  }

  if (AllowReassociate) {
    Value *X, *Y;
    Value *NewInst = nullptr;
    Instruction *SimpR = nullptr;

    if (Op0->hasOneUse() && match(Op0, m_FDiv(m_Value(X), m_Value(Y)))) {
      // (X/Y) / Z => X / (Y*Z)
      //
      if (!isa<Constant>(Y) || !isa<Constant>(Op1)) {
        NewInst = Builder->CreateFMul(Y, Op1);
        if (Instruction *RI = dyn_cast<Instruction>(NewInst)) {
          FastMathFlags Flags = I.getFastMathFlags();
          Flags &= cast<Instruction>(Op0)->getFastMathFlags();
          RI->setFastMathFlags(Flags);
        }
        SimpR = BinaryOperator::CreateFDiv(X, NewInst);
      }
    } else if (Op1->hasOneUse() && match(Op1, m_FDiv(m_Value(X), m_Value(Y)))) {
      // Z / (X/Y) => Z*Y / X
      //
      if (!isa<Constant>(Y) || !isa<Constant>(Op0)) {
        NewInst = Builder->CreateFMul(Op0, Y);
        if (Instruction *RI = dyn_cast<Instruction>(NewInst)) {
          FastMathFlags Flags = I.getFastMathFlags();
          Flags &= cast<Instruction>(Op1)->getFastMathFlags();
          RI->setFastMathFlags(Flags);
        }
        SimpR = BinaryOperator::CreateFDiv(NewInst, X);
      }
    }

    if (NewInst) {
      if (Instruction *T = dyn_cast<Instruction>(NewInst))
        T->setDebugLoc(I.getDebugLoc());
      SimpR->setFastMathFlags(I.getFastMathFlags());
      return SimpR;
    }
  }

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitFenceInst ( FenceInst &  I )

inlineinherited

Definition at line 179 of file InstVisitor.h.

{ DELEGATE(Instruction);}

Instruction * InstCombiner::visitFMul

(

BinaryOperator &

I

)

Definition at line 433 of file InstCombineMulDivRem.cpp.

                                                      {
  bool Changed = SimplifyAssociativeOrCommutative(I);
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (isa<Constant>(Op0))
    std::swap(Op0, Op1);

  if (Value *V = SimplifyFMulInst(Op0, Op1, I.getFastMathFlags(), DL))
    return ReplaceInstUsesWith(I, V);

  bool AllowReassociate = I.hasUnsafeAlgebra();

  // Simplify mul instructions with a constant RHS.
  if (isa<Constant>(Op1)) {
    // Try to fold constant mul into select arguments.
    if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;

    if (isa<PHINode>(Op0))
      if (Instruction *NV = FoldOpIntoPhi(I))
        return NV;

    // (fmul X, -1.0) --> (fsub -0.0, X)
    if (match(Op1, m_SpecificFP(-1.0))) {
      Constant *NegZero = ConstantFP::getNegativeZero(Op1->getType());
      Instruction *RI = BinaryOperator::CreateFSub(NegZero, Op0);
      RI->copyFastMathFlags(&I);
      return RI;
    }

    Constant *C = cast<Constant>(Op1);
    if (AllowReassociate && isFiniteNonZeroFp(C)) {
      // Let MDC denote an expression in one of these forms:
      // X * C, C/X, X/C, where C is a constant.
      //
      // Try to simplify "MDC * Constant"
      if (isFMulOrFDivWithConstant(Op0))
        if (Value *V = foldFMulConst(cast<Instruction>(Op0), C, &I))
          return ReplaceInstUsesWith(I, V);

      // (MDC +/- C1) * C => (MDC * C) +/- (C1 * C)
      Instruction *FAddSub = dyn_cast<Instruction>(Op0);
      if (FAddSub &&
          (FAddSub->getOpcode() == Instruction::FAdd ||
           FAddSub->getOpcode() == Instruction::FSub)) {
        Value *Opnd0 = FAddSub->getOperand(0);
        Value *Opnd1 = FAddSub->getOperand(1);
        Constant *C0 = dyn_cast<Constant>(Opnd0);
        Constant *C1 = dyn_cast<Constant>(Opnd1);
        bool Swap = false;
        if (C0) {
          std::swap(C0, C1);
          std::swap(Opnd0, Opnd1);
          Swap = true;
        }

        if (C1 && isFiniteNonZeroFp(C1) && isFMulOrFDivWithConstant(Opnd0)) {
          Value *M1 = ConstantExpr::getFMul(C1, C);
          Value *M0 = isNormalFp(cast<Constant>(M1)) ?
                      foldFMulConst(cast<Instruction>(Opnd0), C, &I) :
                      nullptr;
          if (M0 && M1) {
            if (Swap && FAddSub->getOpcode() == Instruction::FSub)
              std::swap(M0, M1);

            Instruction *RI = (FAddSub->getOpcode() == Instruction::FAdd)
                                  ? BinaryOperator::CreateFAdd(M0, M1)
                                  : BinaryOperator::CreateFSub(M0, M1);
            RI->copyFastMathFlags(&I);
            return RI;
          }
        }
      }
    }
  }


  // Under unsafe algebra do:
  // X * log2(0.5*Y) = X*log2(Y) - X
  if (I.hasUnsafeAlgebra()) {
    Value *OpX = nullptr;
    Value *OpY = nullptr;
    IntrinsicInst *Log2;
    detectLog2OfHalf(Op0, OpY, Log2);
    if (OpY) {
      OpX = Op1;
    } else {
      detectLog2OfHalf(Op1, OpY, Log2);
      if (OpY) {
        OpX = Op0;
      }
    }
    // if pattern detected emit alternate sequence
    if (OpX && OpY) {
      BuilderTy::FastMathFlagGuard Guard(*Builder);
      Builder->SetFastMathFlags(Log2->getFastMathFlags());
      Log2->setArgOperand(0, OpY);
      Value *FMulVal = Builder->CreateFMul(OpX, Log2);
      Value *FSub = Builder->CreateFSub(FMulVal, OpX);
      FSub->takeName(&I);
      return ReplaceInstUsesWith(I, FSub);
    }
  }

  // Handle symmetric situation in a 2-iteration loop
  Value *Opnd0 = Op0;
  Value *Opnd1 = Op1;
  for (int i = 0; i < 2; i++) {
    bool IgnoreZeroSign = I.hasNoSignedZeros();
    if (BinaryOperator::isFNeg(Opnd0, IgnoreZeroSign)) {
      BuilderTy::FastMathFlagGuard Guard(*Builder);
      Builder->SetFastMathFlags(I.getFastMathFlags());

      Value *N0 = dyn_castFNegVal(Opnd0, IgnoreZeroSign);
      Value *N1 = dyn_castFNegVal(Opnd1, IgnoreZeroSign);

      // -X * -Y => X*Y
      if (N1) {
        Value *FMul = Builder->CreateFMul(N0, N1);
        FMul->takeName(&I);
        return ReplaceInstUsesWith(I, FMul);
      }

      if (Opnd0->hasOneUse()) {
        // -X * Y => -(X*Y) (Promote negation as high as possible)
        Value *T = Builder->CreateFMul(N0, Opnd1);
        Value *Neg = Builder->CreateFNeg(T);
        Neg->takeName(&I);
        return ReplaceInstUsesWith(I, Neg);
      }
    }

    // (X*Y) * X => (X*X) * Y where Y != X
    //  The purpose is two-fold:
    //   1) to form a power expression (of X).
    //   2) potentially shorten the critical path: After transformation, the
    //  latency of the instruction Y is amortized by the expression of X*X,
    //  and therefore Y is in a "less critical" position compared to what it
    //  was before the transformation.
    //
    if (AllowReassociate) {
      Value *Opnd0_0, *Opnd0_1;
      if (Opnd0->hasOneUse() &&
          match(Opnd0, m_FMul(m_Value(Opnd0_0), m_Value(Opnd0_1)))) {
        Value *Y = nullptr;
        if (Opnd0_0 == Opnd1 && Opnd0_1 != Opnd1)
          Y = Opnd0_1;
        else if (Opnd0_1 == Opnd1 && Opnd0_0 != Opnd1)
          Y = Opnd0_0;

        if (Y) {
          BuilderTy::FastMathFlagGuard Guard(*Builder);
          Builder->SetFastMathFlags(I.getFastMathFlags());
          Value *T = Builder->CreateFMul(Opnd1, Opnd1);

          Value *R = Builder->CreateFMul(T, Y);
          R->takeName(&I);
          return ReplaceInstUsesWith(I, R);
        }
      }
    }

    // B * (uitofp i1 C) -> select C, B, 0
    if (I.hasNoNaNs() && I.hasNoInfs() && I.hasNoSignedZeros()) {
      Value *LHS = Op0, *RHS = Op1;
      Value *B, *C;
      if (!match(RHS, m_UIToFP(m_Value(C))))
        std::swap(LHS, RHS);

      if (match(RHS, m_UIToFP(m_Value(C))) &&
          C->getType()->getScalarType()->isIntegerTy(1)) {
        B = LHS;
        Value *Zero = ConstantFP::getNegativeZero(B->getType());
        return SelectInst::Create(C, B, Zero);
      }
    }

    // A * (1 - uitofp i1 C) -> select C, 0, A
    if (I.hasNoNaNs() && I.hasNoInfs() && I.hasNoSignedZeros()) {
      Value *LHS = Op0, *RHS = Op1;
      Value *A, *C;
      if (!match(RHS, m_FSub(m_FPOne(), m_UIToFP(m_Value(C)))))
        std::swap(LHS, RHS);

      if (match(RHS, m_FSub(m_FPOne(), m_UIToFP(m_Value(C)))) &&
          C->getType()->getScalarType()->isIntegerTy(1)) {
        A = LHS;
        Value *Zero = ConstantFP::getNegativeZero(A->getType());
        return SelectInst::Create(C, Zero, A);
      }
    }

    if (!isa<Constant>(Op1))
      std::swap(Opnd0, Opnd1);
    else
      break;
  }

  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitFPExt

(

CastInst &

CI

)

Definition at line 1354 of file InstCombineCasts.cpp.

                                                  {
  return commonCastTransforms(CI);
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitFPExtInst ( FPExtInst &  I )

inlineinherited

Definition at line 186 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitFPToSI

(

FPToSIInst &

FI

)

Definition at line 1378 of file InstCombineCasts.cpp.

                                                     {
  Instruction *OpI = dyn_cast<Instruction>(FI.getOperand(0));
  if (!OpI)
    return commonCastTransforms(FI);

  // fptosi(sitofp(X)) --> X
  // fptosi(uitofp(X)) --> X
  // This is safe if the intermediate type has enough bits in its mantissa to
  // accurately represent all values of X.  For example, do not do this with
  // i64->float->i64.  This is also safe for sitofp case, because any negative
  // 'X' value would cause an undefined result for the fptoui.
  if ((isa<UIToFPInst>(OpI) || isa<SIToFPInst>(OpI)) &&
      OpI->getOperand(0)->getType() == FI.getType() &&
      (int)FI.getType()->getScalarSizeInBits() <=
                    OpI->getType()->getFPMantissaWidth())
    return ReplaceInstUsesWith(FI, OpI->getOperand(0));

  return commonCastTransforms(FI);
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitFPToSIInst ( FPToSIInst &  I )

inlineinherited

Definition at line 188 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitFPToUI

(

FPToUIInst &

FI

)

Definition at line 1358 of file InstCombineCasts.cpp.

                                                     {
  Instruction *OpI = dyn_cast<Instruction>(FI.getOperand(0));
  if (!OpI)
    return commonCastTransforms(FI);

  // fptoui(uitofp(X)) --> X
  // fptoui(sitofp(X)) --> X
  // This is safe if the intermediate type has enough bits in its mantissa to
  // accurately represent all values of X.  For example, do not do this with
  // i64->float->i64.  This is also safe for sitofp case, because any negative
  // 'X' value would cause an undefined result for the fptoui.
  if ((isa<UIToFPInst>(OpI) || isa<SIToFPInst>(OpI)) &&
      OpI->getOperand(0)->getType() == FI.getType() &&
      (int)FI.getType()->getScalarSizeInBits() < /*extra bit for sign */
                    OpI->getType()->getFPMantissaWidth())
    return ReplaceInstUsesWith(FI, OpI->getOperand(0));

  return commonCastTransforms(FI);
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitFPToUIInst ( FPToUIInst &  I )

inlineinherited

Definition at line 187 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitFPTrunc

(

FPTruncInst &

CI

)

Definition at line 1171 of file InstCombineCasts.cpp.

                                                       {
  if (Instruction *I = commonCastTransforms(CI))
    return I;
  // If we have fptrunc(OpI (fpextend x), (fpextend y)), we would like to
  // simpilify this expression to avoid one or more of the trunc/extend
  // operations if we can do so without changing the numerical results.
  //
  // The exact manner in which the widths of the operands interact to limit
  // what we can and cannot do safely varies from operation to operation, and
  // is explained below in the various case statements.
  BinaryOperator *OpI = dyn_cast<BinaryOperator>(CI.getOperand(0));
  if (OpI && OpI->hasOneUse()) {
    Value *LHSOrig = LookThroughFPExtensions(OpI->getOperand(0));
    Value *RHSOrig = LookThroughFPExtensions(OpI->getOperand(1));
    unsigned OpWidth = OpI->getType()->getFPMantissaWidth();
    unsigned LHSWidth = LHSOrig->getType()->getFPMantissaWidth();
    unsigned RHSWidth = RHSOrig->getType()->getFPMantissaWidth();
    unsigned SrcWidth = std::max(LHSWidth, RHSWidth);
    unsigned DstWidth = CI.getType()->getFPMantissaWidth();
    switch (OpI->getOpcode()) {
      default: break;
      case Instruction::FAdd:
      case Instruction::FSub:
        // For addition and subtraction, the infinitely precise result can
        // essentially be arbitrarily wide; proving that double rounding
        // will not occur because the result of OpI is exact (as we will for
        // FMul, for example) is hopeless.  However, we *can* nonetheless
        // frequently know that double rounding cannot occur (or that it is
        // innocuous) by taking advantage of the specific structure of
        // infinitely-precise results that admit double rounding.
        //
        // Specifically, if OpWidth >= 2*DstWdith+1 and DstWidth is sufficient
        // to represent both sources, we can guarantee that the double
        // rounding is innocuous (See p50 of Figueroa's 2000 PhD thesis,
        // "A Rigorous Framework for Fully Supporting the IEEE Standard ..."
        // for proof of this fact).
        //
        // Note: Figueroa does not consider the case where DstFormat !=
        // SrcFormat.  It's possible (likely even!) that this analysis
        // could be tightened for those cases, but they are rare (the main
        // case of interest here is (float)((double)float + float)).
        if (OpWidth >= 2*DstWidth+1 && DstWidth >= SrcWidth) {
          if (LHSOrig->getType() != CI.getType())
            LHSOrig = Builder->CreateFPExt(LHSOrig, CI.getType());
          if (RHSOrig->getType() != CI.getType())
            RHSOrig = Builder->CreateFPExt(RHSOrig, CI.getType());
          Instruction *RI =
            BinaryOperator::Create(OpI->getOpcode(), LHSOrig, RHSOrig);
          RI->copyFastMathFlags(OpI);
          return RI;
        }
        break;
      case Instruction::FMul:
        // For multiplication, the infinitely precise result has at most
        // LHSWidth + RHSWidth significant bits; if OpWidth is sufficient
        // that such a value can be exactly represented, then no double
        // rounding can possibly occur; we can safely perform the operation
        // in the destination format if it can represent both sources.
        if (OpWidth >= LHSWidth + RHSWidth && DstWidth >= SrcWidth) {
          if (LHSOrig->getType() != CI.getType())
            LHSOrig = Builder->CreateFPExt(LHSOrig, CI.getType());
          if (RHSOrig->getType() != CI.getType())
            RHSOrig = Builder->CreateFPExt(RHSOrig, CI.getType());
          Instruction *RI =
            BinaryOperator::CreateFMul(LHSOrig, RHSOrig);
          RI->copyFastMathFlags(OpI);
          return RI;
        }
        break;
      case Instruction::FDiv:
        // For division, we use again use the bound from Figueroa's
        // dissertation.  I am entirely certain that this bound can be
        // tightened in the unbalanced operand case by an analysis based on
        // the diophantine rational approximation bound, but the well-known
        // condition used here is a good conservative first pass.
        // TODO: Tighten bound via rigorous analysis of the unbalanced case.
        if (OpWidth >= 2*DstWidth && DstWidth >= SrcWidth) {
          if (LHSOrig->getType() != CI.getType())
            LHSOrig = Builder->CreateFPExt(LHSOrig, CI.getType());
          if (RHSOrig->getType() != CI.getType())
            RHSOrig = Builder->CreateFPExt(RHSOrig, CI.getType());
          Instruction *RI =
            BinaryOperator::CreateFDiv(LHSOrig, RHSOrig);
          RI->copyFastMathFlags(OpI);
          return RI;
        }
        break;
      case Instruction::FRem:
        // Remainder is straightforward.  Remainder is always exact, so the
        // type of OpI doesn't enter into things at all.  We simply evaluate
        // in whichever source type is larger, then convert to the
        // destination type.
        if (LHSWidth < SrcWidth)
          LHSOrig = Builder->CreateFPExt(LHSOrig, RHSOrig->getType());
        else if (RHSWidth <= SrcWidth)
          RHSOrig = Builder->CreateFPExt(RHSOrig, LHSOrig->getType());
        Value *ExactResult = Builder->CreateFRem(LHSOrig, RHSOrig);
        if (Instruction *RI = dyn_cast<Instruction>(ExactResult))
          RI->copyFastMathFlags(OpI);
        return CastInst::CreateFPCast(ExactResult, CI.getType());
    }

    // (fptrunc (fneg x)) -> (fneg (fptrunc x))
    if (BinaryOperator::isFNeg(OpI)) {
      Value *InnerTrunc = Builder->CreateFPTrunc(OpI->getOperand(1),
                                                 CI.getType());
      Instruction *RI = BinaryOperator::CreateFNeg(InnerTrunc);
      RI->copyFastMathFlags(OpI);
      return RI;
    }
  }

  // (fptrunc (select cond, R1, Cst)) -->
  // (select cond, (fptrunc R1), (fptrunc Cst))
  SelectInst *SI = dyn_cast<SelectInst>(CI.getOperand(0));
  if (SI &&
      (isa<ConstantFP>(SI->getOperand(1)) ||
       isa<ConstantFP>(SI->getOperand(2)))) {
    Value *LHSTrunc = Builder->CreateFPTrunc(SI->getOperand(1),
                                             CI.getType());
    Value *RHSTrunc = Builder->CreateFPTrunc(SI->getOperand(2),
                                             CI.getType());
    return SelectInst::Create(SI->getOperand(0), LHSTrunc, RHSTrunc);
  }

  IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI.getOperand(0));
  if (II) {
    switch (II->getIntrinsicID()) {
      default: break;
      case Intrinsic::fabs: {
        // (fptrunc (fabs x)) -> (fabs (fptrunc x))
        Value *InnerTrunc = Builder->CreateFPTrunc(II->getArgOperand(0),
                                                   CI.getType());
        Type *IntrinsicType[] = { CI.getType() };
        Function *Overload =
          Intrinsic::getDeclaration(CI.getParent()->getParent()->getParent(),
                                    II->getIntrinsicID(), IntrinsicType);

        Value *Args[] = { InnerTrunc };
        return CallInst::Create(Overload, Args, II->getName());
      }
    }
  }

  // Fold (fptrunc (sqrt (fpext x))) -> (sqrtf x)
  // Note that we restrict this transformation based on
  // TLI->has(LibFunc::sqrtf), even for the sqrt intrinsic, because
  // TLI->has(LibFunc::sqrtf) is sufficient to guarantee that the
  // single-precision intrinsic can be expanded in the backend.
  CallInst *Call = dyn_cast<CallInst>(CI.getOperand(0));
  if (Call && Call->getCalledFunction() && TLI->has(LibFunc::sqrtf) &&
      (Call->getCalledFunction()->getName() == TLI->getName(LibFunc::sqrt) ||
       Call->getCalledFunction()->getIntrinsicID() == Intrinsic::sqrt) &&
      Call->getNumArgOperands() == 1 &&
      Call->hasOneUse()) {
    CastInst *Arg = dyn_cast<CastInst>(Call->getArgOperand(0));
    if (Arg && Arg->getOpcode() == Instruction::FPExt &&
        CI.getType()->isFloatTy() &&
        Call->getType()->isDoubleTy() &&
        Arg->getType()->isDoubleTy() &&
        Arg->getOperand(0)->getType()->isFloatTy()) {
      Function *Callee = Call->getCalledFunction();
      Module *M = CI.getParent()->getParent()->getParent();
      Constant *SqrtfFunc = (Callee->getIntrinsicID() == Intrinsic::sqrt) ?
        Intrinsic::getDeclaration(M, Intrinsic::sqrt, Builder->getFloatTy()) :
        M->getOrInsertFunction("sqrtf", Callee->getAttributes(),
                               Builder->getFloatTy(), Builder->getFloatTy(),
                               NULL);
      CallInst *ret = CallInst::Create(SqrtfFunc, Arg->getOperand(0),
                                       "sqrtfcall");
      ret->setAttributes(Callee->getAttributes());


      // Remove the old Call.  With -fmath-errno, it won't get marked readnone.
      ReplaceInstUsesWith(*Call, UndefValue::get(Call->getType()));
      EraseInstFromFunction(*Call);
      return ret;
    }
  }

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitFPTruncInst ( FPTruncInst &  I )

inlineinherited

Definition at line 185 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitFree

(

CallInst &

FI

)

Definition at line 1896 of file InstructionCombining.cpp.

                                                 {
  Value *Op = FI.getArgOperand(0);

  // free undef -> unreachable.
  if (isa<UndefValue>(Op)) {
    // Insert a new store to null because we cannot modify the CFG here.
    Builder->CreateStore(ConstantInt::getTrue(FI.getContext()),
                         UndefValue::get(Type::getInt1PtrTy(FI.getContext())));
    return EraseInstFromFunction(FI);
  }

  // If we have 'free null' delete the instruction.  This can happen in stl code
  // when lots of inlining happens.
  if (isa<ConstantPointerNull>(Op))
    return EraseInstFromFunction(FI);

  // If we optimize for code size, try to move the call to free before the null
  // test so that simplify cfg can remove the empty block and dead code
  // elimination the branch. I.e., helps to turn something like:
  // if (foo) free(foo);
  // into
  // free(foo);
  if (MinimizeSize)
    if (Instruction *I = tryToMoveFreeBeforeNullTest(FI))
      return I;

  return nullptr;
}

Instruction * InstCombiner::visitFRem

(

BinaryOperator &

I

)

Definition at line 1335 of file InstCombineMulDivRem.cpp.

                                                      {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyFRemInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // Handle cases involving: rem X, (select Cond, Y, Z)
  if (isa<SelectInst>(Op1) && SimplifyDivRemOfSelect(I))
    return &I;

  return nullptr;
}

Instruction * InstCombiner::visitFSub

(

BinaryOperator &

I

)

Definition at line 1601 of file InstCombineAddSub.cpp.

                                                      {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyFSubInst(Op0, Op1, I.getFastMathFlags(), DL))
    return ReplaceInstUsesWith(I, V);

  if (isa<Constant>(Op0))
    if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
      if (Instruction *NV = FoldOpIntoSelect(I, SI))
        return NV;

  // If this is a 'B = x-(-A)', change to B = x+A, potentially looking
  // through FP extensions/truncations along the way.
  if (Value *V = dyn_castFNegVal(Op1)) {
    Instruction *NewI = BinaryOperator::CreateFAdd(Op0, V);
    NewI->copyFastMathFlags(&I);
    return NewI;
  }
  if (FPTruncInst *FPTI = dyn_cast<FPTruncInst>(Op1)) {
    if (Value *V = dyn_castFNegVal(FPTI->getOperand(0))) {
      Value *NewTrunc = Builder->CreateFPTrunc(V, I.getType());
      Instruction *NewI = BinaryOperator::CreateFAdd(Op0, NewTrunc);
      NewI->copyFastMathFlags(&I);
      return NewI;
    }
  } else if (FPExtInst *FPEI = dyn_cast<FPExtInst>(Op1)) {
    if (Value *V = dyn_castFNegVal(FPEI->getOperand(0))) {
      Value *NewExt = Builder->CreateFPExt(V, I.getType());
      Instruction *NewI = BinaryOperator::CreateFAdd(Op0, NewExt);
      NewI->copyFastMathFlags(&I);
      return NewI;
    }
  }

  if (I.hasUnsafeAlgebra()) {
    if (Value *V = FAddCombine(Builder).simplify(&I))
      return ReplaceInstUsesWith(I, V);
  }

  return nullptr;
}

void llvm::InstVisitor< InstCombiner , Instruction * >::visitFunction ( Function &  F )

inlineinherited

Definition at line 142 of file InstVisitor.h.

{}

Instruction * InstCombiner::visitGetElementPtrInst

(

GetElementPtrInst &

GEP

)

See if we can simplify: X = bitcast A* to B* Y = gep X, <...constant indices...> into a gep of the original struct. This is important for SROA and alias analysis of unions. If "A" is also a bitcast, wait for A/X to be merged.

Definition at line 1284 of file InstructionCombining.cpp.

                                                                        {
  SmallVector<Value*, 8> Ops(GEP.op_begin(), GEP.op_end());

  if (Value *V = SimplifyGEPInst(Ops, DL))
    return ReplaceInstUsesWith(GEP, V);

  Value *PtrOp = GEP.getOperand(0);

  // Eliminate unneeded casts for indices, and replace indices which displace
  // by multiples of a zero size type with zero.
  if (DL) {
    bool MadeChange = false;
    Type *IntPtrTy = DL->getIntPtrType(GEP.getPointerOperandType());

    gep_type_iterator GTI = gep_type_begin(GEP);
    for (User::op_iterator I = GEP.op_begin() + 1, E = GEP.op_end();
         I != E; ++I, ++GTI) {
      // Skip indices into struct types.
      SequentialType *SeqTy = dyn_cast<SequentialType>(*GTI);
      if (!SeqTy) continue;

      // If the element type has zero size then any index over it is equivalent
      // to an index of zero, so replace it with zero if it is not zero already.
      if (SeqTy->getElementType()->isSized() &&
          DL->getTypeAllocSize(SeqTy->getElementType()) == 0)
        if (!isa<Constant>(*I) || !cast<Constant>(*I)->isNullValue()) {
          *I = Constant::getNullValue(IntPtrTy);
          MadeChange = true;
        }

      Type *IndexTy = (*I)->getType();
      if (IndexTy != IntPtrTy) {
        // If we are using a wider index than needed for this platform, shrink
        // it to what we need.  If narrower, sign-extend it to what we need.
        // This explicit cast can make subsequent optimizations more obvious.
        *I = Builder->CreateIntCast(*I, IntPtrTy, true);
        MadeChange = true;
      }
    }
    if (MadeChange) return &GEP;
  }

  // Check to see if the inputs to the PHI node are getelementptr instructions.
  if (PHINode *PN = dyn_cast<PHINode>(PtrOp)) {
    GetElementPtrInst *Op1 = dyn_cast<GetElementPtrInst>(PN->getOperand(0));
    if (!Op1)
      return nullptr;

    signed DI = -1;

    for (auto I = PN->op_begin()+1, E = PN->op_end(); I !=E; ++I) {
      GetElementPtrInst *Op2 = dyn_cast<GetElementPtrInst>(*I);
      if (!Op2 || Op1->getNumOperands() != Op2->getNumOperands())
        return nullptr;

      // Keep track of the type as we walk the GEP.
      Type *CurTy = Op1->getOperand(0)->getType()->getScalarType();

      for (unsigned J = 0, F = Op1->getNumOperands(); J != F; ++J) {
        if (Op1->getOperand(J)->getType() != Op2->getOperand(J)->getType())
          return nullptr;

        if (Op1->getOperand(J) != Op2->getOperand(J)) {
          if (DI == -1) {
            // We have not seen any differences yet in the GEPs feeding the
            // PHI yet, so we record this one if it is allowed to be a
            // variable.

            // The first two arguments can vary for any GEP, the rest have to be
            // static for struct slots
            if (J > 1 && CurTy->isStructTy())
              return nullptr;

            DI = J;
          } else {
            // The GEP is different by more than one input. While this could be
            // extended to support GEPs that vary by more than one variable it
            // doesn't make sense since it greatly increases the complexity and
            // would result in an R+R+R addressing mode which no backend
            // directly supports and would need to be broken into several
            // simpler instructions anyway.
            return nullptr;
          }
        }

        // Sink down a layer of the type for the next iteration.
        if (J > 0) {
          if (CompositeType *CT = dyn_cast<CompositeType>(CurTy)) {
            CurTy = CT->getTypeAtIndex(Op1->getOperand(J));
          } else {
            CurTy = nullptr;
          }
        }
      }
    }

    GetElementPtrInst *NewGEP = cast<GetElementPtrInst>(Op1->clone());

    if (DI == -1) {
      // All the GEPs feeding the PHI are identical. Clone one down into our
      // BB so that it can be merged with the current GEP.
      GEP.getParent()->getInstList().insert(GEP.getParent()->getFirstNonPHI(),
                                            NewGEP);
    } else {
      // All the GEPs feeding the PHI differ at a single offset. Clone a GEP
      // into the current block so it can be merged, and create a new PHI to
      // set that index.
      Instruction *InsertPt = Builder->GetInsertPoint();
      Builder->SetInsertPoint(PN);
      PHINode *NewPN = Builder->CreatePHI(Op1->getOperand(DI)->getType(),
                                          PN->getNumOperands());
      Builder->SetInsertPoint(InsertPt);

      for (auto &I : PN->operands())
        NewPN->addIncoming(cast<GEPOperator>(I)->getOperand(DI),
                           PN->getIncomingBlock(I));

      NewGEP->setOperand(DI, NewPN);
      GEP.getParent()->getInstList().insert(GEP.getParent()->getFirstNonPHI(),
                                            NewGEP);
      NewGEP->setOperand(DI, NewPN);
    }

    GEP.setOperand(0, NewGEP);
    PtrOp = NewGEP;
  }

  // Combine Indices - If the source pointer to this getelementptr instruction
  // is a getelementptr instruction, combine the indices of the two
  // getelementptr instructions into a single instruction.
  //
  if (GEPOperator *Src = dyn_cast<GEPOperator>(PtrOp)) {
    if (!shouldMergeGEPs(*cast<GEPOperator>(&GEP), *Src))
      return nullptr;

    // Note that if our source is a gep chain itself then we wait for that
    // chain to be resolved before we perform this transformation.  This
    // avoids us creating a TON of code in some cases.
    if (GEPOperator *SrcGEP =
          dyn_cast<GEPOperator>(Src->getOperand(0)))
      if (SrcGEP->getNumOperands() == 2 && shouldMergeGEPs(*Src, *SrcGEP))
        return nullptr;   // Wait until our source is folded to completion.

    SmallVector<Value*, 8> Indices;

    // Find out whether the last index in the source GEP is a sequential idx.
    bool EndsWithSequential = false;
    for (gep_type_iterator I = gep_type_begin(*Src), E = gep_type_end(*Src);
         I != E; ++I)
      EndsWithSequential = !(*I)->isStructTy();

    // Can we combine the two pointer arithmetics offsets?
    if (EndsWithSequential) {
      // Replace: gep (gep %P, long B), long A, ...
      // With:    T = long A+B; gep %P, T, ...
      //
      Value *Sum;
      Value *SO1 = Src->getOperand(Src->getNumOperands()-1);
      Value *GO1 = GEP.getOperand(1);
      if (SO1 == Constant::getNullValue(SO1->getType())) {
        Sum = GO1;
      } else if (GO1 == Constant::getNullValue(GO1->getType())) {
        Sum = SO1;
      } else {
        // If they aren't the same type, then the input hasn't been processed
        // by the loop above yet (which canonicalizes sequential index types to
        // intptr_t).  Just avoid transforming this until the input has been
        // normalized.
        if (SO1->getType() != GO1->getType())
          return nullptr;
        Sum = Builder->CreateAdd(SO1, GO1, PtrOp->getName()+".sum");
      }

      // Update the GEP in place if possible.
      if (Src->getNumOperands() == 2) {
        GEP.setOperand(0, Src->getOperand(0));
        GEP.setOperand(1, Sum);
        return &GEP;
      }
      Indices.append(Src->op_begin()+1, Src->op_end()-1);
      Indices.push_back(Sum);
      Indices.append(GEP.op_begin()+2, GEP.op_end());
    } else if (isa<Constant>(*GEP.idx_begin()) &&
               cast<Constant>(*GEP.idx_begin())->isNullValue() &&
               Src->getNumOperands() != 1) {
      // Otherwise we can do the fold if the first index of the GEP is a zero
      Indices.append(Src->op_begin()+1, Src->op_end());
      Indices.append(GEP.idx_begin()+1, GEP.idx_end());
    }

    if (!Indices.empty())
      return (GEP.isInBounds() && Src->isInBounds()) ?
        GetElementPtrInst::CreateInBounds(Src->getOperand(0), Indices,
                                          GEP.getName()) :
        GetElementPtrInst::Create(Src->getOperand(0), Indices, GEP.getName());
  }

  // Canonicalize (gep i8* X, -(ptrtoint Y)) to (sub (ptrtoint X), (ptrtoint Y))
  // The GEP pattern is emitted by the SCEV expander for certain kinds of
  // pointer arithmetic.
  if (DL && GEP.getNumIndices() == 1 &&
      match(GEP.getOperand(1), m_Neg(m_PtrToInt(m_Value())))) {
    unsigned AS = GEP.getPointerAddressSpace();
    if (GEP.getType() == Builder->getInt8PtrTy(AS) &&
        GEP.getOperand(1)->getType()->getScalarSizeInBits() ==
        DL->getPointerSizeInBits(AS)) {
      Operator *Index = cast<Operator>(GEP.getOperand(1));
      Value *PtrToInt = Builder->CreatePtrToInt(PtrOp, Index->getType());
      Value *NewSub = Builder->CreateSub(PtrToInt, Index->getOperand(1));
      return CastInst::Create(Instruction::IntToPtr, NewSub, GEP.getType());
    }
  }

  // Handle gep(bitcast x) and gep(gep x, 0, 0, 0).
  Value *StrippedPtr = PtrOp->stripPointerCasts();
  PointerType *StrippedPtrTy = dyn_cast<PointerType>(StrippedPtr->getType());

  // We do not handle pointer-vector geps here.
  if (!StrippedPtrTy)
    return nullptr;

  if (StrippedPtr != PtrOp) {
    bool HasZeroPointerIndex = false;
    if (ConstantInt *C = dyn_cast<ConstantInt>(GEP.getOperand(1)))
      HasZeroPointerIndex = C->isZero();

    // Transform: GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ...
    // into     : GEP [10 x i8]* X, i32 0, ...
    //
    // Likewise, transform: GEP (bitcast i8* X to [0 x i8]*), i32 0, ...
    //           into     : GEP i8* X, ...
    //
    // This occurs when the program declares an array extern like "int X[];"
    if (HasZeroPointerIndex) {
      PointerType *CPTy = cast<PointerType>(PtrOp->getType());
      if (ArrayType *CATy =
          dyn_cast<ArrayType>(CPTy->getElementType())) {
        // GEP (bitcast i8* X to [0 x i8]*), i32 0, ... ?
        if (CATy->getElementType() == StrippedPtrTy->getElementType()) {
          // -> GEP i8* X, ...
          SmallVector<Value*, 8> Idx(GEP.idx_begin()+1, GEP.idx_end());
          GetElementPtrInst *Res =
            GetElementPtrInst::Create(StrippedPtr, Idx, GEP.getName());
          Res->setIsInBounds(GEP.isInBounds());
          if (StrippedPtrTy->getAddressSpace() == GEP.getAddressSpace())
            return Res;
          // Insert Res, and create an addrspacecast.
          // e.g.,
          // GEP (addrspacecast i8 addrspace(1)* X to [0 x i8]*), i32 0, ...
          // ->
          // %0 = GEP i8 addrspace(1)* X, ...
          // addrspacecast i8 addrspace(1)* %0 to i8*
          return new AddrSpaceCastInst(Builder->Insert(Res), GEP.getType());
        }

        if (ArrayType *XATy =
              dyn_cast<ArrayType>(StrippedPtrTy->getElementType())){
          // GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ... ?
          if (CATy->getElementType() == XATy->getElementType()) {
            // -> GEP [10 x i8]* X, i32 0, ...
            // At this point, we know that the cast source type is a pointer
            // to an array of the same type as the destination pointer
            // array.  Because the array type is never stepped over (there
            // is a leading zero) we can fold the cast into this GEP.
            if (StrippedPtrTy->getAddressSpace() == GEP.getAddressSpace()) {
              GEP.setOperand(0, StrippedPtr);
              return &GEP;
            }
            // Cannot replace the base pointer directly because StrippedPtr's
            // address space is different. Instead, create a new GEP followed by
            // an addrspacecast.
            // e.g.,
            // GEP (addrspacecast [10 x i8] addrspace(1)* X to [0 x i8]*),
            //   i32 0, ...
            // ->
            // %0 = GEP [10 x i8] addrspace(1)* X, ...
            // addrspacecast i8 addrspace(1)* %0 to i8*
            SmallVector<Value*, 8> Idx(GEP.idx_begin(), GEP.idx_end());
            Value *NewGEP = GEP.isInBounds() ?
              Builder->CreateInBoundsGEP(StrippedPtr, Idx, GEP.getName()) :
              Builder->CreateGEP(StrippedPtr, Idx, GEP.getName());
            return new AddrSpaceCastInst(NewGEP, GEP.getType());
          }
        }
      }
    } else if (GEP.getNumOperands() == 2) {
      // Transform things like:
      // %t = getelementptr i32* bitcast ([2 x i32]* %str to i32*), i32 %V
      // into:  %t1 = getelementptr [2 x i32]* %str, i32 0, i32 %V; bitcast
      Type *SrcElTy = StrippedPtrTy->getElementType();
      Type *ResElTy = PtrOp->getType()->getPointerElementType();
      if (DL && SrcElTy->isArrayTy() &&
          DL->getTypeAllocSize(SrcElTy->getArrayElementType()) ==
          DL->getTypeAllocSize(ResElTy)) {
        Type *IdxType = DL->getIntPtrType(GEP.getType());
        Value *Idx[2] = { Constant::getNullValue(IdxType), GEP.getOperand(1) };
        Value *NewGEP = GEP.isInBounds() ?
          Builder->CreateInBoundsGEP(StrippedPtr, Idx, GEP.getName()) :
          Builder->CreateGEP(StrippedPtr, Idx, GEP.getName());

        // V and GEP are both pointer types --> BitCast
        return CastInst::CreatePointerBitCastOrAddrSpaceCast(NewGEP,
                                                             GEP.getType());
      }

      // Transform things like:
      // %V = mul i64 %N, 4
      // %t = getelementptr i8* bitcast (i32* %arr to i8*), i32 %V
      // into:  %t1 = getelementptr i32* %arr, i32 %N; bitcast
      if (DL && ResElTy->isSized() && SrcElTy->isSized()) {
        // Check that changing the type amounts to dividing the index by a scale
        // factor.
        uint64_t ResSize = DL->getTypeAllocSize(ResElTy);
        uint64_t SrcSize = DL->getTypeAllocSize(SrcElTy);
        if (ResSize && SrcSize % ResSize == 0) {
          Value *Idx = GEP.getOperand(1);
          unsigned BitWidth = Idx->getType()->getPrimitiveSizeInBits();
          uint64_t Scale = SrcSize / ResSize;

          // Earlier transforms ensure that the index has type IntPtrType, which
          // considerably simplifies the logic by eliminating implicit casts.
          assert(Idx->getType() == DL->getIntPtrType(GEP.getType()) &&
                 "Index not cast to pointer width?");

          bool NSW;
          if (Value *NewIdx = Descale(Idx, APInt(BitWidth, Scale), NSW)) {
            // Successfully decomposed Idx as NewIdx * Scale, form a new GEP.
            // If the multiplication NewIdx * Scale may overflow then the new
            // GEP may not be "inbounds".
            Value *NewGEP = GEP.isInBounds() && NSW ?
              Builder->CreateInBoundsGEP(StrippedPtr, NewIdx, GEP.getName()) :
              Builder->CreateGEP(StrippedPtr, NewIdx, GEP.getName());

            // The NewGEP must be pointer typed, so must the old one -> BitCast
            return CastInst::CreatePointerBitCastOrAddrSpaceCast(NewGEP,
                                                                 GEP.getType());
          }
        }
      }

      // Similarly, transform things like:
      // getelementptr i8* bitcast ([100 x double]* X to i8*), i32 %tmp
      //   (where tmp = 8*tmp2) into:
      // getelementptr [100 x double]* %arr, i32 0, i32 %tmp2; bitcast
      if (DL && ResElTy->isSized() && SrcElTy->isSized() &&
          SrcElTy->isArrayTy()) {
        // Check that changing to the array element type amounts to dividing the
        // index by a scale factor.
        uint64_t ResSize = DL->getTypeAllocSize(ResElTy);
        uint64_t ArrayEltSize
          = DL->getTypeAllocSize(SrcElTy->getArrayElementType());
        if (ResSize && ArrayEltSize % ResSize == 0) {
          Value *Idx = GEP.getOperand(1);
          unsigned BitWidth = Idx->getType()->getPrimitiveSizeInBits();
          uint64_t Scale = ArrayEltSize / ResSize;

          // Earlier transforms ensure that the index has type IntPtrType, which
          // considerably simplifies the logic by eliminating implicit casts.
          assert(Idx->getType() == DL->getIntPtrType(GEP.getType()) &&
                 "Index not cast to pointer width?");

          bool NSW;
          if (Value *NewIdx = Descale(Idx, APInt(BitWidth, Scale), NSW)) {
            // Successfully decomposed Idx as NewIdx * Scale, form a new GEP.
            // If the multiplication NewIdx * Scale may overflow then the new
            // GEP may not be "inbounds".
            Value *Off[2] = {
              Constant::getNullValue(DL->getIntPtrType(GEP.getType())),
              NewIdx
            };

            Value *NewGEP = GEP.isInBounds() && NSW ?
              Builder->CreateInBoundsGEP(StrippedPtr, Off, GEP.getName()) :
              Builder->CreateGEP(StrippedPtr, Off, GEP.getName());
            // The NewGEP must be pointer typed, so must the old one -> BitCast
            return CastInst::CreatePointerBitCastOrAddrSpaceCast(NewGEP,
                                                                 GEP.getType());
          }
        }
      }
    }
  }

  if (!DL)
    return nullptr;

  /// See if we can simplify:
  ///   X = bitcast A* to B*
  ///   Y = gep X, <...constant indices...>
  /// into a gep of the original struct.  This is important for SROA and alias
  /// analysis of unions.  If "A" is also a bitcast, wait for A/X to be merged.
  if (BitCastInst *BCI = dyn_cast<BitCastInst>(PtrOp)) {
    Value *Operand = BCI->getOperand(0);
    PointerType *OpType = cast<PointerType>(Operand->getType());
    unsigned OffsetBits = DL->getPointerTypeSizeInBits(OpType);
    APInt Offset(OffsetBits, 0);
    if (!isa<BitCastInst>(Operand) &&
        GEP.accumulateConstantOffset(*DL, Offset) &&
        StrippedPtrTy->getAddressSpace() == GEP.getPointerAddressSpace()) {

      // If this GEP instruction doesn't move the pointer, just replace the GEP
      // with a bitcast of the real input to the dest type.
      if (!Offset) {
        // If the bitcast is of an allocation, and the allocation will be
        // converted to match the type of the cast, don't touch this.
        if (isa<AllocaInst>(Operand) || isAllocationFn(Operand, TLI)) {
          // See if the bitcast simplifies, if so, don't nuke this GEP yet.
          if (Instruction *I = visitBitCast(*BCI)) {
            if (I != BCI) {
              I->takeName(BCI);
              BCI->getParent()->getInstList().insert(BCI, I);
              ReplaceInstUsesWith(*BCI, I);
            }
            return &GEP;
          }
        }
        return new BitCastInst(Operand, GEP.getType());
      }

      // Otherwise, if the offset is non-zero, we need to find out if there is a
      // field at Offset in 'A's type.  If so, we can pull the cast through the
      // GEP.
      SmallVector<Value*, 8> NewIndices;
      if (FindElementAtOffset(OpType, Offset.getSExtValue(), NewIndices)) {
        Value *NGEP = GEP.isInBounds() ?
          Builder->CreateInBoundsGEP(Operand, NewIndices) :
          Builder->CreateGEP(Operand, NewIndices);

        if (NGEP->getType() == GEP.getType())
          return ReplaceInstUsesWith(GEP, NGEP);
        NGEP->takeName(&GEP);
        return new BitCastInst(NGEP, GEP.getType());
      }
    }
  }

  return nullptr;
}

Instruction * InstCombiner::visitICmpInst

(

ICmpInst &

I

)

Orders the operands of the compare so that they are listed from most complex to least complex. This puts constants before unary operators, before binary operators.

Definition at line 2327 of file InstCombineCompares.cpp.

                                                    {
  bool Changed = false;
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
  unsigned Op0Cplxity = getComplexity(Op0);
  unsigned Op1Cplxity = getComplexity(Op1);

  /// Orders the operands of the compare so that they are listed from most
  /// complex to least complex.  This puts constants before unary operators,
  /// before binary operators.
  if (Op0Cplxity < Op1Cplxity ||
        (Op0Cplxity == Op1Cplxity &&
         swapMayExposeCSEOpportunities(Op0, Op1))) {
    I.swapOperands();
    std::swap(Op0, Op1);
    Changed = true;
  }

  if (Value *V = SimplifyICmpInst(I.getPredicate(), Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // comparing -val or val with non-zero is the same as just comparing val
  // ie, abs(val) != 0 -> val != 0
  if (I.getPredicate() == ICmpInst::ICMP_NE && match(Op1, m_Zero()))
  {
    Value *Cond, *SelectTrue, *SelectFalse;
    if (match(Op0, m_Select(m_Value(Cond), m_Value(SelectTrue),
                            m_Value(SelectFalse)))) {
      if (Value *V = dyn_castNegVal(SelectTrue)) {
        if (V == SelectFalse)
          return CmpInst::Create(Instruction::ICmp, I.getPredicate(), V, Op1);
      }
      else if (Value *V = dyn_castNegVal(SelectFalse)) {
        if (V == SelectTrue)
          return CmpInst::Create(Instruction::ICmp, I.getPredicate(), V, Op1);
      }
    }
  }

  Type *Ty = Op0->getType();

  // icmp's with boolean values can always be turned into bitwise operations
  if (Ty->isIntegerTy(1)) {
    switch (I.getPredicate()) {
    default: llvm_unreachable("Invalid icmp instruction!");
    case ICmpInst::ICMP_EQ: {               // icmp eq i1 A, B -> ~(A^B)
      Value *Xor = Builder->CreateXor(Op0, Op1, I.getName()+"tmp");
      return BinaryOperator::CreateNot(Xor);
    }
    case ICmpInst::ICMP_NE:                  // icmp eq i1 A, B -> A^B
      return BinaryOperator::CreateXor(Op0, Op1);

    case ICmpInst::ICMP_UGT:
      std::swap(Op0, Op1);                   // Change icmp ugt -> icmp ult
      // FALL THROUGH
    case ICmpInst::ICMP_ULT:{               // icmp ult i1 A, B -> ~A & B
      Value *Not = Builder->CreateNot(Op0, I.getName()+"tmp");
      return BinaryOperator::CreateAnd(Not, Op1);
    }
    case ICmpInst::ICMP_SGT:
      std::swap(Op0, Op1);                   // Change icmp sgt -> icmp slt
      // FALL THROUGH
    case ICmpInst::ICMP_SLT: {               // icmp slt i1 A, B -> A & ~B
      Value *Not = Builder->CreateNot(Op1, I.getName()+"tmp");
      return BinaryOperator::CreateAnd(Not, Op0);
    }
    case ICmpInst::ICMP_UGE:
      std::swap(Op0, Op1);                   // Change icmp uge -> icmp ule
      // FALL THROUGH
    case ICmpInst::ICMP_ULE: {               //  icmp ule i1 A, B -> ~A | B
      Value *Not = Builder->CreateNot(Op0, I.getName()+"tmp");
      return BinaryOperator::CreateOr(Not, Op1);
    }
    case ICmpInst::ICMP_SGE:
      std::swap(Op0, Op1);                   // Change icmp sge -> icmp sle
      // FALL THROUGH
    case ICmpInst::ICMP_SLE: {               //  icmp sle i1 A, B -> A | ~B
      Value *Not = Builder->CreateNot(Op1, I.getName()+"tmp");
      return BinaryOperator::CreateOr(Not, Op0);
    }
    }
  }

  unsigned BitWidth = 0;
  if (Ty->isIntOrIntVectorTy())
    BitWidth = Ty->getScalarSizeInBits();
  else if (DL)  // Pointers require DL info to get their size.
    BitWidth = DL->getTypeSizeInBits(Ty->getScalarType());

  bool isSignBit = false;

  // See if we are doing a comparison with a constant.
  if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
    Value *A = nullptr, *B = nullptr;

    // Match the following pattern, which is a common idiom when writing
    // overflow-safe integer arithmetic function.  The source performs an
    // addition in wider type, and explicitly checks for overflow using
    // comparisons against INT_MIN and INT_MAX.  Simplify this by using the
    // sadd_with_overflow intrinsic.
    //
    // TODO: This could probably be generalized to handle other overflow-safe
    // operations if we worked out the formulas to compute the appropriate
    // magic constants.
    //
    // sum = a + b
    // if (sum+128 >u 255)  ...  -> llvm.sadd.with.overflow.i8
    {
    ConstantInt *CI2;    // I = icmp ugt (add (add A, B), CI2), CI
    if (I.getPredicate() == ICmpInst::ICMP_UGT &&
        match(Op0, m_Add(m_Add(m_Value(A), m_Value(B)), m_ConstantInt(CI2))))
      if (Instruction *Res = ProcessUGT_ADDCST_ADD(I, A, B, CI2, CI, *this))
        return Res;
    }

    // (icmp ne/eq (sub A B) 0) -> (icmp ne/eq A, B)
    if (I.isEquality() && CI->isZero() &&
        match(Op0, m_Sub(m_Value(A), m_Value(B)))) {
      // (icmp cond A B) if cond is equality
      return new ICmpInst(I.getPredicate(), A, B);
    }

    // If we have an icmp le or icmp ge instruction, turn it into the
    // appropriate icmp lt or icmp gt instruction.  This allows us to rely on
    // them being folded in the code below.  The SimplifyICmpInst code has
    // already handled the edge cases for us, so we just assert on them.
    switch (I.getPredicate()) {
    default: break;
    case ICmpInst::ICMP_ULE:
      assert(!CI->isMaxValue(false));                 // A <=u MAX -> TRUE
      return new ICmpInst(ICmpInst::ICMP_ULT, Op0,
                          Builder->getInt(CI->getValue()+1));
    case ICmpInst::ICMP_SLE:
      assert(!CI->isMaxValue(true));                  // A <=s MAX -> TRUE
      return new ICmpInst(ICmpInst::ICMP_SLT, Op0,
                          Builder->getInt(CI->getValue()+1));
    case ICmpInst::ICMP_UGE:
      assert(!CI->isMinValue(false));                 // A >=u MIN -> TRUE
      return new ICmpInst(ICmpInst::ICMP_UGT, Op0,
                          Builder->getInt(CI->getValue()-1));
    case ICmpInst::ICMP_SGE:
      assert(!CI->isMinValue(true));                  // A >=s MIN -> TRUE
      return new ICmpInst(ICmpInst::ICMP_SGT, Op0,
                          Builder->getInt(CI->getValue()-1));
    }

    // If this comparison is a normal comparison, it demands all
    // bits, if it is a sign bit comparison, it only demands the sign bit.
    bool UnusedBit;
    isSignBit = isSignBitCheck(I.getPredicate(), CI, UnusedBit);
  }

  // See if we can fold the comparison based on range information we can get
  // by checking whether bits are known to be zero or one in the input.
  if (BitWidth != 0) {
    APInt Op0KnownZero(BitWidth, 0), Op0KnownOne(BitWidth, 0);
    APInt Op1KnownZero(BitWidth, 0), Op1KnownOne(BitWidth, 0);

    if (SimplifyDemandedBits(I.getOperandUse(0),
                             DemandedBitsLHSMask(I, BitWidth, isSignBit),
                             Op0KnownZero, Op0KnownOne, 0))
      return &I;
    if (SimplifyDemandedBits(I.getOperandUse(1),
                             APInt::getAllOnesValue(BitWidth),
                             Op1KnownZero, Op1KnownOne, 0))
      return &I;

    // Given the known and unknown bits, compute a range that the LHS could be
    // in.  Compute the Min, Max and RHS values based on the known bits. For the
    // EQ and NE we use unsigned values.
    APInt Op0Min(BitWidth, 0), Op0Max(BitWidth, 0);
    APInt Op1Min(BitWidth, 0), Op1Max(BitWidth, 0);
    if (I.isSigned()) {
      ComputeSignedMinMaxValuesFromKnownBits(Op0KnownZero, Op0KnownOne,
                                             Op0Min, Op0Max);
      ComputeSignedMinMaxValuesFromKnownBits(Op1KnownZero, Op1KnownOne,
                                             Op1Min, Op1Max);
    } else {
      ComputeUnsignedMinMaxValuesFromKnownBits(Op0KnownZero, Op0KnownOne,
                                               Op0Min, Op0Max);
      ComputeUnsignedMinMaxValuesFromKnownBits(Op1KnownZero, Op1KnownOne,
                                               Op1Min, Op1Max);
    }

    // If Min and Max are known to be the same, then SimplifyDemandedBits
    // figured out that the LHS is a constant.  Just constant fold this now so
    // that code below can assume that Min != Max.
    if (!isa<Constant>(Op0) && Op0Min == Op0Max)
      return new ICmpInst(I.getPredicate(),
                          ConstantInt::get(Op0->getType(), Op0Min), Op1);
    if (!isa<Constant>(Op1) && Op1Min == Op1Max)
      return new ICmpInst(I.getPredicate(), Op0,
                          ConstantInt::get(Op1->getType(), Op1Min));

    // Based on the range information we know about the LHS, see if we can
    // simplify this comparison.  For example, (x&4) < 8 is always true.
    switch (I.getPredicate()) {
    default: llvm_unreachable("Unknown icmp opcode!");
    case ICmpInst::ICMP_EQ: {
      if (Op0Max.ult(Op1Min) || Op0Min.ugt(Op1Max))
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));

      // If all bits are known zero except for one, then we know at most one
      // bit is set.   If the comparison is against zero, then this is a check
      // to see if *that* bit is set.
      APInt Op0KnownZeroInverted = ~Op0KnownZero;
      if (~Op1KnownZero == 0) {
        // If the LHS is an AND with the same constant, look through it.
        Value *LHS = nullptr;
        ConstantInt *LHSC = nullptr;
        if (!match(Op0, m_And(m_Value(LHS), m_ConstantInt(LHSC))) ||
            LHSC->getValue() != Op0KnownZeroInverted)
          LHS = Op0;

        // If the LHS is 1 << x, and we know the result is a power of 2 like 8,
        // then turn "((1 << x)&8) == 0" into "x != 3".
        // or turn "((1 << x)&7) == 0" into "x > 2".
        Value *X = nullptr;
        if (match(LHS, m_Shl(m_One(), m_Value(X)))) {
          APInt ValToCheck = Op0KnownZeroInverted;
          if (ValToCheck.isPowerOf2()) {
            unsigned CmpVal = ValToCheck.countTrailingZeros();
            return new ICmpInst(ICmpInst::ICMP_NE, X,
                                ConstantInt::get(X->getType(), CmpVal));
          } else if ((++ValToCheck).isPowerOf2()) {
            unsigned CmpVal = ValToCheck.countTrailingZeros() - 1;
            return new ICmpInst(ICmpInst::ICMP_UGT, X,
                                ConstantInt::get(X->getType(), CmpVal));
          }
        }

        // If the LHS is 8 >>u x, and we know the result is a power of 2 like 1,
        // then turn "((8 >>u x)&1) == 0" into "x != 3".
        const APInt *CI;
        if (Op0KnownZeroInverted == 1 &&
            match(LHS, m_LShr(m_Power2(CI), m_Value(X))))
          return new ICmpInst(ICmpInst::ICMP_NE, X,
                              ConstantInt::get(X->getType(),
                                               CI->countTrailingZeros()));
      }

      break;
    }
    case ICmpInst::ICMP_NE: {
      if (Op0Max.ult(Op1Min) || Op0Min.ugt(Op1Max))
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));

      // If all bits are known zero except for one, then we know at most one
      // bit is set.   If the comparison is against zero, then this is a check
      // to see if *that* bit is set.
      APInt Op0KnownZeroInverted = ~Op0KnownZero;
      if (~Op1KnownZero == 0) {
        // If the LHS is an AND with the same constant, look through it.
        Value *LHS = nullptr;
        ConstantInt *LHSC = nullptr;
        if (!match(Op0, m_And(m_Value(LHS), m_ConstantInt(LHSC))) ||
            LHSC->getValue() != Op0KnownZeroInverted)
          LHS = Op0;

        // If the LHS is 1 << x, and we know the result is a power of 2 like 8,
        // then turn "((1 << x)&8) != 0" into "x == 3".
        // or turn "((1 << x)&7) != 0" into "x < 3".
        Value *X = nullptr;
        if (match(LHS, m_Shl(m_One(), m_Value(X)))) {
          APInt ValToCheck = Op0KnownZeroInverted;
          if (ValToCheck.isPowerOf2()) {
            unsigned CmpVal = ValToCheck.countTrailingZeros();
            return new ICmpInst(ICmpInst::ICMP_EQ, X,
                                ConstantInt::get(X->getType(), CmpVal));
          } else if ((++ValToCheck).isPowerOf2()) {
            unsigned CmpVal = ValToCheck.countTrailingZeros();
            return new ICmpInst(ICmpInst::ICMP_ULT, X,
                                ConstantInt::get(X->getType(), CmpVal));
          }
        }

        // If the LHS is 8 >>u x, and we know the result is a power of 2 like 1,
        // then turn "((8 >>u x)&1) != 0" into "x == 3".
        const APInt *CI;
        if (Op0KnownZeroInverted == 1 &&
            match(LHS, m_LShr(m_Power2(CI), m_Value(X))))
          return new ICmpInst(ICmpInst::ICMP_EQ, X,
                              ConstantInt::get(X->getType(),
                                               CI->countTrailingZeros()));
      }

      break;
    }
    case ICmpInst::ICMP_ULT:
      if (Op0Max.ult(Op1Min))          // A <u B -> true if max(A) < min(B)
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
      if (Op0Min.uge(Op1Max))          // A <u B -> false if min(A) >= max(B)
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));
      if (Op1Min == Op0Max)            // A <u B -> A != B if max(A) == min(B)
        return new ICmpInst(ICmpInst::ICMP_NE, Op0, Op1);
      if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
        if (Op1Max == Op0Min+1)        // A <u C -> A == C-1 if min(A)+1 == C
          return new ICmpInst(ICmpInst::ICMP_EQ, Op0,
                              Builder->getInt(CI->getValue()-1));

        // (x <u 2147483648) -> (x >s -1)  -> true if sign bit clear
        if (CI->isMinValue(true))
          return new ICmpInst(ICmpInst::ICMP_SGT, Op0,
                           Constant::getAllOnesValue(Op0->getType()));
      }
      break;
    case ICmpInst::ICMP_UGT:
      if (Op0Min.ugt(Op1Max))          // A >u B -> true if min(A) > max(B)
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
      if (Op0Max.ule(Op1Min))          // A >u B -> false if max(A) <= max(B)
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));

      if (Op1Max == Op0Min)            // A >u B -> A != B if min(A) == max(B)
        return new ICmpInst(ICmpInst::ICMP_NE, Op0, Op1);
      if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
        if (Op1Min == Op0Max-1)        // A >u C -> A == C+1 if max(a)-1 == C
          return new ICmpInst(ICmpInst::ICMP_EQ, Op0,
                              Builder->getInt(CI->getValue()+1));

        // (x >u 2147483647) -> (x <s 0)  -> true if sign bit set
        if (CI->isMaxValue(true))
          return new ICmpInst(ICmpInst::ICMP_SLT, Op0,
                              Constant::getNullValue(Op0->getType()));
      }
      break;
    case ICmpInst::ICMP_SLT:
      if (Op0Max.slt(Op1Min))          // A <s B -> true if max(A) < min(C)
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
      if (Op0Min.sge(Op1Max))          // A <s B -> false if min(A) >= max(C)
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));
      if (Op1Min == Op0Max)            // A <s B -> A != B if max(A) == min(B)
        return new ICmpInst(ICmpInst::ICMP_NE, Op0, Op1);
      if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
        if (Op1Max == Op0Min+1)        // A <s C -> A == C-1 if min(A)+1 == C
          return new ICmpInst(ICmpInst::ICMP_EQ, Op0,
                              Builder->getInt(CI->getValue()-1));
      }
      break;
    case ICmpInst::ICMP_SGT:
      if (Op0Min.sgt(Op1Max))          // A >s B -> true if min(A) > max(B)
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
      if (Op0Max.sle(Op1Min))          // A >s B -> false if max(A) <= min(B)
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));

      if (Op1Max == Op0Min)            // A >s B -> A != B if min(A) == max(B)
        return new ICmpInst(ICmpInst::ICMP_NE, Op0, Op1);
      if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
        if (Op1Min == Op0Max-1)        // A >s C -> A == C+1 if max(A)-1 == C
          return new ICmpInst(ICmpInst::ICMP_EQ, Op0,
                              Builder->getInt(CI->getValue()+1));
      }
      break;
    case ICmpInst::ICMP_SGE:
      assert(!isa<ConstantInt>(Op1) && "ICMP_SGE with ConstantInt not folded!");
      if (Op0Min.sge(Op1Max))          // A >=s B -> true if min(A) >= max(B)
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
      if (Op0Max.slt(Op1Min))          // A >=s B -> false if max(A) < min(B)
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));
      break;
    case ICmpInst::ICMP_SLE:
      assert(!isa<ConstantInt>(Op1) && "ICMP_SLE with ConstantInt not folded!");
      if (Op0Max.sle(Op1Min))          // A <=s B -> true if max(A) <= min(B)
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
      if (Op0Min.sgt(Op1Max))          // A <=s B -> false if min(A) > max(B)
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));
      break;
    case ICmpInst::ICMP_UGE:
      assert(!isa<ConstantInt>(Op1) && "ICMP_UGE with ConstantInt not folded!");
      if (Op0Min.uge(Op1Max))          // A >=u B -> true if min(A) >= max(B)
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
      if (Op0Max.ult(Op1Min))          // A >=u B -> false if max(A) < min(B)
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));
      break;
    case ICmpInst::ICMP_ULE:
      assert(!isa<ConstantInt>(Op1) && "ICMP_ULE with ConstantInt not folded!");
      if (Op0Max.ule(Op1Min))          // A <=u B -> true if max(A) <= min(B)
        return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
      if (Op0Min.ugt(Op1Max))          // A <=u B -> false if min(A) > max(B)
        return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));
      break;
    }

    // Turn a signed comparison into an unsigned one if both operands
    // are known to have the same sign.
    if (I.isSigned() &&
        ((Op0KnownZero.isNegative() && Op1KnownZero.isNegative()) ||
         (Op0KnownOne.isNegative() && Op1KnownOne.isNegative())))
      return new ICmpInst(I.getUnsignedPredicate(), Op0, Op1);
  }

  // Test if the ICmpInst instruction is used exclusively by a select as
  // part of a minimum or maximum operation. If so, refrain from doing
  // any other folding. This helps out other analyses which understand
  // non-obfuscated minimum and maximum idioms, such as ScalarEvolution
  // and CodeGen. And in this case, at least one of the comparison
  // operands has at least one user besides the compare (the select),
  // which would often largely negate the benefit of folding anyway.
  if (I.hasOneUse())
    if (SelectInst *SI = dyn_cast<SelectInst>(*I.user_begin()))
      if ((SI->getOperand(1) == Op0 && SI->getOperand(2) == Op1) ||
          (SI->getOperand(2) == Op0 && SI->getOperand(1) == Op1))
        return nullptr;

  // See if we are doing a comparison between a constant and an instruction that
  // can be folded into the comparison.
  if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
    // Since the RHS is a ConstantInt (CI), if the left hand side is an
    // instruction, see if that instruction also has constants so that the
    // instruction can be folded into the icmp
    if (Instruction *LHSI = dyn_cast<Instruction>(Op0))
      if (Instruction *Res = visitICmpInstWithInstAndIntCst(I, LHSI, CI))
        return Res;
  }

  // Handle icmp with constant (but not simple integer constant) RHS
  if (Constant *RHSC = dyn_cast<Constant>(Op1)) {
    if (Instruction *LHSI = dyn_cast<Instruction>(Op0))
      switch (LHSI->getOpcode()) {
      case Instruction::GetElementPtr:
          // icmp pred GEP (P, int 0, int 0, int 0), null -> icmp pred P, null
        if (RHSC->isNullValue() &&
            cast<GetElementPtrInst>(LHSI)->hasAllZeroIndices())
          return new ICmpInst(I.getPredicate(), LHSI->getOperand(0),
                  Constant::getNullValue(LHSI->getOperand(0)->getType()));
        break;
      case Instruction::PHI:
        // Only fold icmp into the PHI if the phi and icmp are in the same
        // block.  If in the same block, we're encouraging jump threading.  If
        // not, we are just pessimizing the code by making an i1 phi.
        if (LHSI->getParent() == I.getParent())
          if (Instruction *NV = FoldOpIntoPhi(I))
            return NV;
        break;
      case Instruction::Select: {
        // If either operand of the select is a constant, we can fold the
        // comparison into the select arms, which will cause one to be
        // constant folded and the select turned into a bitwise or.
        Value *Op1 = nullptr, *Op2 = nullptr;
        if (Constant *C = dyn_cast<Constant>(LHSI->getOperand(1)))
          Op1 = ConstantExpr::getICmp(I.getPredicate(), C, RHSC);
        if (Constant *C = dyn_cast<Constant>(LHSI->getOperand(2)))
          Op2 = ConstantExpr::getICmp(I.getPredicate(), C, RHSC);

        // We only want to perform this transformation if it will not lead to
        // additional code. This is true if either both sides of the select
        // fold to a constant (in which case the icmp is replaced with a select
        // which will usually simplify) or this is the only user of the
        // select (in which case we are trading a select+icmp for a simpler
        // select+icmp).
        if ((Op1 && Op2) || (LHSI->hasOneUse() && (Op1 || Op2))) {
          if (!Op1)
            Op1 = Builder->CreateICmp(I.getPredicate(), LHSI->getOperand(1),
                                      RHSC, I.getName());
          if (!Op2)
            Op2 = Builder->CreateICmp(I.getPredicate(), LHSI->getOperand(2),
                                      RHSC, I.getName());
          return SelectInst::Create(LHSI->getOperand(0), Op1, Op2);
        }
        break;
      }
      case Instruction::IntToPtr:
        // icmp pred inttoptr(X), null -> icmp pred X, 0
        if (RHSC->isNullValue() && DL &&
            DL->getIntPtrType(RHSC->getType()) ==
               LHSI->getOperand(0)->getType())
          return new ICmpInst(I.getPredicate(), LHSI->getOperand(0),
                        Constant::getNullValue(LHSI->getOperand(0)->getType()));
        break;

      case Instruction::Load:
        // Try to optimize things like "A[i] > 4" to index computations.
        if (GetElementPtrInst *GEP =
              dyn_cast<GetElementPtrInst>(LHSI->getOperand(0))) {
          if (GlobalVariable *GV = dyn_cast<GlobalVariable>(GEP->getOperand(0)))
            if (GV->isConstant() && GV->hasDefinitiveInitializer() &&
                !cast<LoadInst>(LHSI)->isVolatile())
              if (Instruction *Res = FoldCmpLoadFromIndexedGlobal(GEP, GV, I))
                return Res;
        }
        break;
      }
  }

  // If we can optimize a 'icmp GEP, P' or 'icmp P, GEP', do so now.
  if (GEPOperator *GEP = dyn_cast<GEPOperator>(Op0))
    if (Instruction *NI = FoldGEPICmp(GEP, Op1, I.getPredicate(), I))
      return NI;
  if (GEPOperator *GEP = dyn_cast<GEPOperator>(Op1))
    if (Instruction *NI = FoldGEPICmp(GEP, Op0,
                           ICmpInst::getSwappedPredicate(I.getPredicate()), I))
      return NI;

  // Test to see if the operands of the icmp are casted versions of other
  // values.  If the ptr->ptr cast can be stripped off both arguments, we do so
  // now.
  if (BitCastInst *CI = dyn_cast<BitCastInst>(Op0)) {
    if (Op0->getType()->isPointerTy() &&
        (isa<Constant>(Op1) || isa<BitCastInst>(Op1))) {
      // We keep moving the cast from the left operand over to the right
      // operand, where it can often be eliminated completely.
      Op0 = CI->getOperand(0);

      // If operand #1 is a bitcast instruction, it must also be a ptr->ptr cast
      // so eliminate it as well.
      if (BitCastInst *CI2 = dyn_cast<BitCastInst>(Op1))
        Op1 = CI2->getOperand(0);

      // If Op1 is a constant, we can fold the cast into the constant.
      if (Op0->getType() != Op1->getType()) {
        if (Constant *Op1C = dyn_cast<Constant>(Op1)) {
          Op1 = ConstantExpr::getBitCast(Op1C, Op0->getType());
        } else {
          // Otherwise, cast the RHS right before the icmp
          Op1 = Builder->CreateBitCast(Op1, Op0->getType());
        }
      }
      return new ICmpInst(I.getPredicate(), Op0, Op1);
    }
  }

  if (isa<CastInst>(Op0)) {
    // Handle the special case of: icmp (cast bool to X), <cst>
    // This comes up when you have code like
    //   int X = A < B;
    //   if (X) ...
    // For generality, we handle any zero-extension of any operand comparison
    // with a constant or another cast from the same type.
    if (isa<Constant>(Op1) || isa<CastInst>(Op1))
      if (Instruction *R = visitICmpInstWithCastAndCast(I))
        return R;
  }

  // Special logic for binary operators.
  BinaryOperator *BO0 = dyn_cast<BinaryOperator>(Op0);
  BinaryOperator *BO1 = dyn_cast<BinaryOperator>(Op1);
  if (BO0 || BO1) {
    CmpInst::Predicate Pred = I.getPredicate();
    bool NoOp0WrapProblem = false, NoOp1WrapProblem = false;
    if (BO0 && isa<OverflowingBinaryOperator>(BO0))
      NoOp0WrapProblem = ICmpInst::isEquality(Pred) ||
        (CmpInst::isUnsigned(Pred) && BO0->hasNoUnsignedWrap()) ||
        (CmpInst::isSigned(Pred) && BO0->hasNoSignedWrap());
    if (BO1 && isa<OverflowingBinaryOperator>(BO1))
      NoOp1WrapProblem = ICmpInst::isEquality(Pred) ||
        (CmpInst::isUnsigned(Pred) && BO1->hasNoUnsignedWrap()) ||
        (CmpInst::isSigned(Pred) && BO1->hasNoSignedWrap());

    // Analyze the case when either Op0 or Op1 is an add instruction.
    // Op0 = A + B (or A and B are null); Op1 = C + D (or C and D are null).
    Value *A = nullptr, *B = nullptr, *C = nullptr, *D = nullptr;
    if (BO0 && BO0->getOpcode() == Instruction::Add)
      A = BO0->getOperand(0), B = BO0->getOperand(1);
    if (BO1 && BO1->getOpcode() == Instruction::Add)
      C = BO1->getOperand(0), D = BO1->getOperand(1);

    // icmp (X+Y), X -> icmp Y, 0 for equalities or if there is no overflow.
    if ((A == Op1 || B == Op1) && NoOp0WrapProblem)
      return new ICmpInst(Pred, A == Op1 ? B : A,
                          Constant::getNullValue(Op1->getType()));

    // icmp X, (X+Y) -> icmp 0, Y for equalities or if there is no overflow.
    if ((C == Op0 || D == Op0) && NoOp1WrapProblem)
      return new ICmpInst(Pred, Constant::getNullValue(Op0->getType()),
                          C == Op0 ? D : C);

    // icmp (X+Y), (X+Z) -> icmp Y, Z for equalities or if there is no overflow.
    if (A && C && (A == C || A == D || B == C || B == D) &&
        NoOp0WrapProblem && NoOp1WrapProblem &&
        // Try not to increase register pressure.
        BO0->hasOneUse() && BO1->hasOneUse()) {
      // Determine Y and Z in the form icmp (X+Y), (X+Z).
      Value *Y, *Z;
      if (A == C) {
        // C + B == C + D  ->  B == D
        Y = B;
        Z = D;
      } else if (A == D) {
        // D + B == C + D  ->  B == C
        Y = B;
        Z = C;
      } else if (B == C) {
        // A + C == C + D  ->  A == D
        Y = A;
        Z = D;
      } else {
        assert(B == D);
        // A + D == C + D  ->  A == C
        Y = A;
        Z = C;
      }
      return new ICmpInst(Pred, Y, Z);
    }

    // icmp slt (X + -1), Y -> icmp sle X, Y
    if (A && NoOp0WrapProblem && Pred == CmpInst::ICMP_SLT &&
        match(B, m_AllOnes()))
      return new ICmpInst(CmpInst::ICMP_SLE, A, Op1);

    // icmp sge (X + -1), Y -> icmp sgt X, Y
    if (A && NoOp0WrapProblem && Pred == CmpInst::ICMP_SGE &&
        match(B, m_AllOnes()))
      return new ICmpInst(CmpInst::ICMP_SGT, A, Op1);

    // icmp sle (X + 1), Y -> icmp slt X, Y
    if (A && NoOp0WrapProblem && Pred == CmpInst::ICMP_SLE &&
        match(B, m_One()))
      return new ICmpInst(CmpInst::ICMP_SLT, A, Op1);

    // icmp sgt (X + 1), Y -> icmp sge X, Y
    if (A && NoOp0WrapProblem && Pred == CmpInst::ICMP_SGT &&
        match(B, m_One()))
      return new ICmpInst(CmpInst::ICMP_SGE, A, Op1);

    // if C1 has greater magnitude than C2:
    //  icmp (X + C1), (Y + C2) -> icmp (X + C3), Y
    //  s.t. C3 = C1 - C2
    //
    // if C2 has greater magnitude than C1:
    //  icmp (X + C1), (Y + C2) -> icmp X, (Y + C3)
    //  s.t. C3 = C2 - C1
    if (A && C && NoOp0WrapProblem && NoOp1WrapProblem &&
        (BO0->hasOneUse() || BO1->hasOneUse()) && !I.isUnsigned())
      if (ConstantInt *C1 = dyn_cast<ConstantInt>(B))
        if (ConstantInt *C2 = dyn_cast<ConstantInt>(D)) {
          const APInt &AP1 = C1->getValue();
          const APInt &AP2 = C2->getValue();
          if (AP1.isNegative() == AP2.isNegative()) {
            APInt AP1Abs = C1->getValue().abs();
            APInt AP2Abs = C2->getValue().abs();
            if (AP1Abs.uge(AP2Abs)) {
              ConstantInt *C3 = Builder->getInt(AP1 - AP2);
              Value *NewAdd = Builder->CreateNSWAdd(A, C3);
              return new ICmpInst(Pred, NewAdd, C);
            } else {
              ConstantInt *C3 = Builder->getInt(AP2 - AP1);
              Value *NewAdd = Builder->CreateNSWAdd(C, C3);
              return new ICmpInst(Pred, A, NewAdd);
            }
          }
        }


    // Analyze the case when either Op0 or Op1 is a sub instruction.
    // Op0 = A - B (or A and B are null); Op1 = C - D (or C and D are null).
    A = nullptr; B = nullptr; C = nullptr; D = nullptr;
    if (BO0 && BO0->getOpcode() == Instruction::Sub)
      A = BO0->getOperand(0), B = BO0->getOperand(1);
    if (BO1 && BO1->getOpcode() == Instruction::Sub)
      C = BO1->getOperand(0), D = BO1->getOperand(1);

    // icmp (X-Y), X -> icmp 0, Y for equalities or if there is no overflow.
    if (A == Op1 && NoOp0WrapProblem)
      return new ICmpInst(Pred, Constant::getNullValue(Op1->getType()), B);

    // icmp X, (X-Y) -> icmp Y, 0 for equalities or if there is no overflow.
    if (C == Op0 && NoOp1WrapProblem)
      return new ICmpInst(Pred, D, Constant::getNullValue(Op0->getType()));

    // icmp (Y-X), (Z-X) -> icmp Y, Z for equalities or if there is no overflow.
    if (B && D && B == D && NoOp0WrapProblem && NoOp1WrapProblem &&
        // Try not to increase register pressure.
        BO0->hasOneUse() && BO1->hasOneUse())
      return new ICmpInst(Pred, A, C);

    // icmp (X-Y), (X-Z) -> icmp Z, Y for equalities or if there is no overflow.
    if (A && C && A == C && NoOp0WrapProblem && NoOp1WrapProblem &&
        // Try not to increase register pressure.
        BO0->hasOneUse() && BO1->hasOneUse())
      return new ICmpInst(Pred, D, B);

    // icmp (0-X) < cst --> x > -cst
    if (NoOp0WrapProblem && ICmpInst::isSigned(Pred)) {
      Value *X;
      if (match(BO0, m_Neg(m_Value(X))))
        if (ConstantInt *RHSC = dyn_cast<ConstantInt>(Op1))
          if (!RHSC->isMinValue(/*isSigned=*/true))
            return new ICmpInst(I.getSwappedPredicate(), X,
                                ConstantExpr::getNeg(RHSC));
    }

    BinaryOperator *SRem = nullptr;
    // icmp (srem X, Y), Y
    if (BO0 && BO0->getOpcode() == Instruction::SRem &&
        Op1 == BO0->getOperand(1))
      SRem = BO0;
    // icmp Y, (srem X, Y)
    else if (BO1 && BO1->getOpcode() == Instruction::SRem &&
             Op0 == BO1->getOperand(1))
      SRem = BO1;
    if (SRem) {
      // We don't check hasOneUse to avoid increasing register pressure because
      // the value we use is the same value this instruction was already using.
      switch (SRem == BO0 ? ICmpInst::getSwappedPredicate(Pred) : Pred) {
        default: break;
        case ICmpInst::ICMP_EQ:
          return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));
        case ICmpInst::ICMP_NE:
          return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
        case ICmpInst::ICMP_SGT:
        case ICmpInst::ICMP_SGE:
          return new ICmpInst(ICmpInst::ICMP_SGT, SRem->getOperand(1),
                              Constant::getAllOnesValue(SRem->getType()));
        case ICmpInst::ICMP_SLT:
        case ICmpInst::ICMP_SLE:
          return new ICmpInst(ICmpInst::ICMP_SLT, SRem->getOperand(1),
                              Constant::getNullValue(SRem->getType()));
      }
    }

    if (BO0 && BO1 && BO0->getOpcode() == BO1->getOpcode() &&
        BO0->hasOneUse() && BO1->hasOneUse() &&
        BO0->getOperand(1) == BO1->getOperand(1)) {
      switch (BO0->getOpcode()) {
      default: break;
      case Instruction::Add:
      case Instruction::Sub:
      case Instruction::Xor:
        if (I.isEquality())    // a+x icmp eq/ne b+x --> a icmp b
          return new ICmpInst(I.getPredicate(), BO0->getOperand(0),
                              BO1->getOperand(0));
        // icmp u/s (a ^ signbit), (b ^ signbit) --> icmp s/u a, b
        if (ConstantInt *CI = dyn_cast<ConstantInt>(BO0->getOperand(1))) {
          if (CI->getValue().isSignBit()) {
            ICmpInst::Predicate Pred = I.isSigned()
                                           ? I.getUnsignedPredicate()
                                           : I.getSignedPredicate();
            return new ICmpInst(Pred, BO0->getOperand(0),
                                BO1->getOperand(0));
          }

          if (CI->isMaxValue(true)) {
            ICmpInst::Predicate Pred = I.isSigned()
                                           ? I.getUnsignedPredicate()
                                           : I.getSignedPredicate();
            Pred = I.getSwappedPredicate(Pred);
            return new ICmpInst(Pred, BO0->getOperand(0),
                                BO1->getOperand(0));
          }
        }
        break;
      case Instruction::Mul:
        if (!I.isEquality())
          break;

        if (ConstantInt *CI = dyn_cast<ConstantInt>(BO0->getOperand(1))) {
          // a * Cst icmp eq/ne b * Cst --> a & Mask icmp b & Mask
          // Mask = -1 >> count-trailing-zeros(Cst).
          if (!CI->isZero() && !CI->isOne()) {
            const APInt &AP = CI->getValue();
            ConstantInt *Mask = ConstantInt::get(I.getContext(),
                                    APInt::getLowBitsSet(AP.getBitWidth(),
                                                         AP.getBitWidth() -
                                                    AP.countTrailingZeros()));
            Value *And1 = Builder->CreateAnd(BO0->getOperand(0), Mask);
            Value *And2 = Builder->CreateAnd(BO1->getOperand(0), Mask);
            return new ICmpInst(I.getPredicate(), And1, And2);
          }
        }
        break;
      case Instruction::UDiv:
      case Instruction::LShr:
        if (I.isSigned())
          break;
        // fall-through
      case Instruction::SDiv:
      case Instruction::AShr:
        if (!BO0->isExact() || !BO1->isExact())
          break;
        return new ICmpInst(I.getPredicate(), BO0->getOperand(0),
                            BO1->getOperand(0));
      case Instruction::Shl: {
        bool NUW = BO0->hasNoUnsignedWrap() && BO1->hasNoUnsignedWrap();
        bool NSW = BO0->hasNoSignedWrap() && BO1->hasNoSignedWrap();
        if (!NUW && !NSW)
          break;
        if (!NSW && I.isSigned())
          break;
        return new ICmpInst(I.getPredicate(), BO0->getOperand(0),
                            BO1->getOperand(0));
      }
      }
    }
  }

  { Value *A, *B;
    // Transform (A & ~B) == 0 --> (A & B) != 0
    // and       (A & ~B) != 0 --> (A & B) == 0
    // if A is a power of 2.
    if (match(Op0, m_And(m_Value(A), m_Not(m_Value(B)))) &&
        match(Op1, m_Zero()) && isKnownToBeAPowerOfTwo(A) && I.isEquality())
      return new ICmpInst(I.getInversePredicate(),
                          Builder->CreateAnd(A, B),
                          Op1);

    // ~x < ~y --> y < x
    // ~x < cst --> ~cst < x
    if (match(Op0, m_Not(m_Value(A)))) {
      if (match(Op1, m_Not(m_Value(B))))
        return new ICmpInst(I.getPredicate(), B, A);
      if (ConstantInt *RHSC = dyn_cast<ConstantInt>(Op1))
        return new ICmpInst(I.getPredicate(), ConstantExpr::getNot(RHSC), A);
    }

    // (a+b) <u a  --> llvm.uadd.with.overflow.
    // (a+b) <u b  --> llvm.uadd.with.overflow.
    if (I.getPredicate() == ICmpInst::ICMP_ULT &&
        match(Op0, m_Add(m_Value(A), m_Value(B))) &&
        (Op1 == A || Op1 == B))
      if (Instruction *R = ProcessUAddIdiom(I, Op0, *this))
        return R;

    // a >u (a+b)  --> llvm.uadd.with.overflow.
    // b >u (a+b)  --> llvm.uadd.with.overflow.
    if (I.getPredicate() == ICmpInst::ICMP_UGT &&
        match(Op1, m_Add(m_Value(A), m_Value(B))) &&
        (Op0 == A || Op0 == B))
      if (Instruction *R = ProcessUAddIdiom(I, Op1, *this))
        return R;

    // (zext a) * (zext b)  --> llvm.umul.with.overflow.
    if (match(Op0, m_Mul(m_ZExt(m_Value(A)), m_ZExt(m_Value(B))))) {
      if (Instruction *R = ProcessUMulZExtIdiom(I, Op0, Op1, *this))
        return R;
    }
    if (match(Op1, m_Mul(m_ZExt(m_Value(A)), m_ZExt(m_Value(B))))) {
      if (Instruction *R = ProcessUMulZExtIdiom(I, Op1, Op0, *this))
        return R;
    }
  }

  if (I.isEquality()) {
    Value *A, *B, *C, *D;

    if (match(Op0, m_Xor(m_Value(A), m_Value(B)))) {
      if (A == Op1 || B == Op1) {    // (A^B) == A  ->  B == 0
        Value *OtherVal = A == Op1 ? B : A;
        return new ICmpInst(I.getPredicate(), OtherVal,
                            Constant::getNullValue(A->getType()));
      }

      if (match(Op1, m_Xor(m_Value(C), m_Value(D)))) {
        // A^c1 == C^c2 --> A == C^(c1^c2)
        ConstantInt *C1, *C2;
        if (match(B, m_ConstantInt(C1)) &&
            match(D, m_ConstantInt(C2)) && Op1->hasOneUse()) {
          Constant *NC = Builder->getInt(C1->getValue() ^ C2->getValue());
          Value *Xor = Builder->CreateXor(C, NC);
          return new ICmpInst(I.getPredicate(), A, Xor);
        }

        // A^B == A^D -> B == D
        if (A == C) return new ICmpInst(I.getPredicate(), B, D);
        if (A == D) return new ICmpInst(I.getPredicate(), B, C);
        if (B == C) return new ICmpInst(I.getPredicate(), A, D);
        if (B == D) return new ICmpInst(I.getPredicate(), A, C);
      }
    }

    if (match(Op1, m_Xor(m_Value(A), m_Value(B))) &&
        (A == Op0 || B == Op0)) {
      // A == (A^B)  ->  B == 0
      Value *OtherVal = A == Op0 ? B : A;
      return new ICmpInst(I.getPredicate(), OtherVal,
                          Constant::getNullValue(A->getType()));
    }

    // (X&Z) == (Y&Z) -> (X^Y) & Z == 0
    if (match(Op0, m_OneUse(m_And(m_Value(A), m_Value(B)))) &&
        match(Op1, m_OneUse(m_And(m_Value(C), m_Value(D))))) {
      Value *X = nullptr, *Y = nullptr, *Z = nullptr;

      if (A == C) {
        X = B; Y = D; Z = A;
      } else if (A == D) {
        X = B; Y = C; Z = A;
      } else if (B == C) {
        X = A; Y = D; Z = B;
      } else if (B == D) {
        X = A; Y = C; Z = B;
      }

      if (X) {   // Build (X^Y) & Z
        Op1 = Builder->CreateXor(X, Y);
        Op1 = Builder->CreateAnd(Op1, Z);
        I.setOperand(0, Op1);
        I.setOperand(1, Constant::getNullValue(Op1->getType()));
        return &I;
      }
    }

    // Transform (zext A) == (B & (1<<X)-1) --> A == (trunc B)
    // and       (B & (1<<X)-1) == (zext A) --> A == (trunc B)
    ConstantInt *Cst1;
    if ((Op0->hasOneUse() &&
         match(Op0, m_ZExt(m_Value(A))) &&
         match(Op1, m_And(m_Value(B), m_ConstantInt(Cst1)))) ||
        (Op1->hasOneUse() &&
         match(Op0, m_And(m_Value(B), m_ConstantInt(Cst1))) &&
         match(Op1, m_ZExt(m_Value(A))))) {
      APInt Pow2 = Cst1->getValue() + 1;
      if (Pow2.isPowerOf2() && isa<IntegerType>(A->getType()) &&
          Pow2.logBase2() == cast<IntegerType>(A->getType())->getBitWidth())
        return new ICmpInst(I.getPredicate(), A,
                            Builder->CreateTrunc(B, A->getType()));
    }

    // (A >> C) == (B >> C) --> (A^B) u< (1 << C)
    // For lshr and ashr pairs.
    if ((match(Op0, m_OneUse(m_LShr(m_Value(A), m_ConstantInt(Cst1)))) &&
         match(Op1, m_OneUse(m_LShr(m_Value(B), m_Specific(Cst1))))) ||
        (match(Op0, m_OneUse(m_AShr(m_Value(A), m_ConstantInt(Cst1)))) &&
         match(Op1, m_OneUse(m_AShr(m_Value(B), m_Specific(Cst1)))))) {
      unsigned TypeBits = Cst1->getBitWidth();
      unsigned ShAmt = (unsigned)Cst1->getLimitedValue(TypeBits);
      if (ShAmt < TypeBits && ShAmt != 0) {
        ICmpInst::Predicate Pred = I.getPredicate() == ICmpInst::ICMP_NE
                                       ? ICmpInst::ICMP_UGE
                                       : ICmpInst::ICMP_ULT;
        Value *Xor = Builder->CreateXor(A, B, I.getName() + ".unshifted");
        APInt CmpVal = APInt::getOneBitSet(TypeBits, ShAmt);
        return new ICmpInst(Pred, Xor, Builder->getInt(CmpVal));
      }
    }

    // Transform "icmp eq (trunc (lshr(X, cst1)), cst" to
    // "icmp (and X, mask), cst"
    uint64_t ShAmt = 0;
    if (Op0->hasOneUse() &&
        match(Op0, m_Trunc(m_OneUse(m_LShr(m_Value(A),
                                           m_ConstantInt(ShAmt))))) &&
        match(Op1, m_ConstantInt(Cst1)) &&
        // Only do this when A has multiple uses.  This is most important to do
        // when it exposes other optimizations.
        !A->hasOneUse()) {
      unsigned ASize =cast<IntegerType>(A->getType())->getPrimitiveSizeInBits();

      if (ShAmt < ASize) {
        APInt MaskV =
          APInt::getLowBitsSet(ASize, Op0->getType()->getPrimitiveSizeInBits());
        MaskV <<= ShAmt;

        APInt CmpV = Cst1->getValue().zext(ASize);
        CmpV <<= ShAmt;

        Value *Mask = Builder->CreateAnd(A, Builder->getInt(MaskV));
        return new ICmpInst(I.getPredicate(), Mask, Builder->getInt(CmpV));
      }
    }
  }

  {
    Value *X; ConstantInt *Cst;
    // icmp X+Cst, X
    if (match(Op0, m_Add(m_Value(X), m_ConstantInt(Cst))) && Op1 == X)
      return FoldICmpAddOpCst(I, X, Cst, I.getPredicate());

    // icmp X, X+Cst
    if (match(Op1, m_Add(m_Value(X), m_ConstantInt(Cst))) && Op0 == X)
      return FoldICmpAddOpCst(I, X, Cst, I.getSwappedPredicate());
  }
  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitICmpInstWithCastAndCast

(

ICmpInst &

ICI

)

visitICmpInstWithCastAndCast - Handle icmp (cast x to y), (cast/cst). We only handle extending casts so far.

Definition at line 1790 of file InstCombineCompares.cpp.

                                                                     {
  const CastInst *LHSCI = cast<CastInst>(ICI.getOperand(0));
  Value *LHSCIOp        = LHSCI->getOperand(0);
  Type *SrcTy     = LHSCIOp->getType();
  Type *DestTy    = LHSCI->getType();
  Value *RHSCIOp;

  // Turn icmp (ptrtoint x), (ptrtoint/c) into a compare of the input if the
  // integer type is the same size as the pointer type.
  if (DL && LHSCI->getOpcode() == Instruction::PtrToInt &&
      DL->getPointerTypeSizeInBits(SrcTy) == DestTy->getIntegerBitWidth()) {
    Value *RHSOp = nullptr;
    if (Constant *RHSC = dyn_cast<Constant>(ICI.getOperand(1))) {
      RHSOp = ConstantExpr::getIntToPtr(RHSC, SrcTy);
    } else if (PtrToIntInst *RHSC = dyn_cast<PtrToIntInst>(ICI.getOperand(1))) {
      RHSOp = RHSC->getOperand(0);
      // If the pointer types don't match, insert a bitcast.
      if (LHSCIOp->getType() != RHSOp->getType())
        RHSOp = Builder->CreateBitCast(RHSOp, LHSCIOp->getType());
    }

    if (RHSOp)
      return new ICmpInst(ICI.getPredicate(), LHSCIOp, RHSOp);
  }

  // The code below only handles extension cast instructions, so far.
  // Enforce this.
  if (LHSCI->getOpcode() != Instruction::ZExt &&
      LHSCI->getOpcode() != Instruction::SExt)
    return nullptr;

  bool isSignedExt = LHSCI->getOpcode() == Instruction::SExt;
  bool isSignedCmp = ICI.isSigned();

  if (CastInst *CI = dyn_cast<CastInst>(ICI.getOperand(1))) {
    // Not an extension from the same type?
    RHSCIOp = CI->getOperand(0);
    if (RHSCIOp->getType() != LHSCIOp->getType())
      return nullptr;

    // If the signedness of the two casts doesn't agree (i.e. one is a sext
    // and the other is a zext), then we can't handle this.
    if (CI->getOpcode() != LHSCI->getOpcode())
      return nullptr;

    // Deal with equality cases early.
    if (ICI.isEquality())
      return new ICmpInst(ICI.getPredicate(), LHSCIOp, RHSCIOp);

    // A signed comparison of sign extended values simplifies into a
    // signed comparison.
    if (isSignedCmp && isSignedExt)
      return new ICmpInst(ICI.getPredicate(), LHSCIOp, RHSCIOp);

    // The other three cases all fold into an unsigned comparison.
    return new ICmpInst(ICI.getUnsignedPredicate(), LHSCIOp, RHSCIOp);
  }

  // If we aren't dealing with a constant on the RHS, exit early
  ConstantInt *CI = dyn_cast<ConstantInt>(ICI.getOperand(1));
  if (!CI)
    return nullptr;

  // Compute the constant that would happen if we truncated to SrcTy then
  // reextended to DestTy.
  Constant *Res1 = ConstantExpr::getTrunc(CI, SrcTy);
  Constant *Res2 = ConstantExpr::getCast(LHSCI->getOpcode(),
                                                Res1, DestTy);

  // If the re-extended constant didn't change...
  if (Res2 == CI) {
    // Deal with equality cases early.
    if (ICI.isEquality())
      return new ICmpInst(ICI.getPredicate(), LHSCIOp, Res1);

    // A signed comparison of sign extended values simplifies into a
    // signed comparison.
    if (isSignedExt && isSignedCmp)
      return new ICmpInst(ICI.getPredicate(), LHSCIOp, Res1);

    // The other three cases all fold into an unsigned comparison.
    return new ICmpInst(ICI.getUnsignedPredicate(), LHSCIOp, Res1);
  }

  // The re-extended constant changed so the constant cannot be represented
  // in the shorter type. Consequently, we cannot emit a simple comparison.
  // All the cases that fold to true or false will have already been handled
  // by SimplifyICmpInst, so only deal with the tricky case.

  if (isSignedCmp || !isSignedExt)
    return nullptr;

  // Evaluate the comparison for LT (we invert for GT below). LE and GE cases
  // should have been folded away previously and not enter in here.

  // We're performing an unsigned comp with a sign extended value.
  // This is true if the input is >= 0. [aka >s -1]
  Constant *NegOne = Constant::getAllOnesValue(SrcTy);
  Value *Result = Builder->CreateICmpSGT(LHSCIOp, NegOne, ICI.getName());

  // Finally, return the value computed.
  if (ICI.getPredicate() == ICmpInst::ICMP_ULT)
    return ReplaceInstUsesWith(ICI, Result);

  assert(ICI.getPredicate() == ICmpInst::ICMP_UGT && "ICmp should be folded!");
  return BinaryOperator::CreateNot(Result);
}

Instruction * InstCombiner::visitICmpInstWithInstAndIntCst	(	ICmpInst &	ICI,
		Instruction *	LHSI,
		ConstantInt *	RHS
	)

visitICmpInstWithInstAndIntCst - Handle "icmp (instr, intcst)".

Definition at line 1050 of file InstCombineCompares.cpp.

                                                                            {
  const APInt &RHSV = RHS->getValue();

  switch (LHSI->getOpcode()) {
  case Instruction::Trunc:
    if (ICI.isEquality() && LHSI->hasOneUse()) {
      // Simplify icmp eq (trunc x to i8), 42 -> icmp eq x, 42|highbits if all
      // of the high bits truncated out of x are known.
      unsigned DstBits = LHSI->getType()->getPrimitiveSizeInBits(),
             SrcBits = LHSI->getOperand(0)->getType()->getPrimitiveSizeInBits();
      APInt KnownZero(SrcBits, 0), KnownOne(SrcBits, 0);
      computeKnownBits(LHSI->getOperand(0), KnownZero, KnownOne);

      // If all the high bits are known, we can do this xform.
      if ((KnownZero|KnownOne).countLeadingOnes() >= SrcBits-DstBits) {
        // Pull in the high bits from known-ones set.
        APInt NewRHS = RHS->getValue().zext(SrcBits);
        NewRHS |= KnownOne & APInt::getHighBitsSet(SrcBits, SrcBits-DstBits);
        return new ICmpInst(ICI.getPredicate(), LHSI->getOperand(0),
                            Builder->getInt(NewRHS));
      }
    }
    break;

  case Instruction::Xor:         // (icmp pred (xor X, XorCst), CI)
    if (ConstantInt *XorCst = dyn_cast<ConstantInt>(LHSI->getOperand(1))) {
      // If this is a comparison that tests the signbit (X < 0) or (x > -1),
      // fold the xor.
      if ((ICI.getPredicate() == ICmpInst::ICMP_SLT && RHSV == 0) ||
          (ICI.getPredicate() == ICmpInst::ICMP_SGT && RHSV.isAllOnesValue())) {
        Value *CompareVal = LHSI->getOperand(0);

        // If the sign bit of the XorCst is not set, there is no change to
        // the operation, just stop using the Xor.
        if (!XorCst->isNegative()) {
          ICI.setOperand(0, CompareVal);
          Worklist.Add(LHSI);
          return &ICI;
        }

        // Was the old condition true if the operand is positive?
        bool isTrueIfPositive = ICI.getPredicate() == ICmpInst::ICMP_SGT;

        // If so, the new one isn't.
        isTrueIfPositive ^= true;

        if (isTrueIfPositive)
          return new ICmpInst(ICmpInst::ICMP_SGT, CompareVal,
                              SubOne(RHS));
        else
          return new ICmpInst(ICmpInst::ICMP_SLT, CompareVal,
                              AddOne(RHS));
      }

      if (LHSI->hasOneUse()) {
        // (icmp u/s (xor A SignBit), C) -> (icmp s/u A, (xor C SignBit))
        if (!ICI.isEquality() && XorCst->getValue().isSignBit()) {
          const APInt &SignBit = XorCst->getValue();
          ICmpInst::Predicate Pred = ICI.isSigned()
                                         ? ICI.getUnsignedPredicate()
                                         : ICI.getSignedPredicate();
          return new ICmpInst(Pred, LHSI->getOperand(0),
                              Builder->getInt(RHSV ^ SignBit));
        }

        // (icmp u/s (xor A ~SignBit), C) -> (icmp s/u (xor C ~SignBit), A)
        if (!ICI.isEquality() && XorCst->isMaxValue(true)) {
          const APInt &NotSignBit = XorCst->getValue();
          ICmpInst::Predicate Pred = ICI.isSigned()
                                         ? ICI.getUnsignedPredicate()
                                         : ICI.getSignedPredicate();
          Pred = ICI.getSwappedPredicate(Pred);
          return new ICmpInst(Pred, LHSI->getOperand(0),
                              Builder->getInt(RHSV ^ NotSignBit));
        }
      }

      // (icmp ugt (xor X, C), ~C) -> (icmp ult X, C)
      //   iff -C is a power of 2
      if (ICI.getPredicate() == ICmpInst::ICMP_UGT &&
          XorCst->getValue() == ~RHSV && (RHSV + 1).isPowerOf2())
        return new ICmpInst(ICmpInst::ICMP_ULT, LHSI->getOperand(0), XorCst);

      // (icmp ult (xor X, C), -C) -> (icmp uge X, C)
      //   iff -C is a power of 2
      if (ICI.getPredicate() == ICmpInst::ICMP_ULT &&
          XorCst->getValue() == -RHSV && RHSV.isPowerOf2())
        return new ICmpInst(ICmpInst::ICMP_UGE, LHSI->getOperand(0), XorCst);
    }
    break;
  case Instruction::And:         // (icmp pred (and X, AndCst), RHS)
    if (LHSI->hasOneUse() && isa<ConstantInt>(LHSI->getOperand(1)) &&
        LHSI->getOperand(0)->hasOneUse()) {
      ConstantInt *AndCst = cast<ConstantInt>(LHSI->getOperand(1));

      // If the LHS is an AND of a truncating cast, we can widen the
      // and/compare to be the input width without changing the value
      // produced, eliminating a cast.
      if (TruncInst *Cast = dyn_cast<TruncInst>(LHSI->getOperand(0))) {
        // We can do this transformation if either the AND constant does not
        // have its sign bit set or if it is an equality comparison.
        // Extending a relational comparison when we're checking the sign
        // bit would not work.
        if (ICI.isEquality() ||
            (!AndCst->isNegative() && RHSV.isNonNegative())) {
          Value *NewAnd =
            Builder->CreateAnd(Cast->getOperand(0),
                               ConstantExpr::getZExt(AndCst, Cast->getSrcTy()));
          NewAnd->takeName(LHSI);
          return new ICmpInst(ICI.getPredicate(), NewAnd,
                              ConstantExpr::getZExt(RHS, Cast->getSrcTy()));
        }
      }

      // If the LHS is an AND of a zext, and we have an equality compare, we can
      // shrink the and/compare to the smaller type, eliminating the cast.
      if (ZExtInst *Cast = dyn_cast<ZExtInst>(LHSI->getOperand(0))) {
        IntegerType *Ty = cast<IntegerType>(Cast->getSrcTy());
        // Make sure we don't compare the upper bits, SimplifyDemandedBits
        // should fold the icmp to true/false in that case.
        if (ICI.isEquality() && RHSV.getActiveBits() <= Ty->getBitWidth()) {
          Value *NewAnd =
            Builder->CreateAnd(Cast->getOperand(0),
                               ConstantExpr::getTrunc(AndCst, Ty));
          NewAnd->takeName(LHSI);
          return new ICmpInst(ICI.getPredicate(), NewAnd,
                              ConstantExpr::getTrunc(RHS, Ty));
        }
      }

      // If this is: (X >> C1) & C2 != C3 (where any shift and any compare
      // could exist), turn it into (X & (C2 << C1)) != (C3 << C1).  This
      // happens a LOT in code produced by the C front-end, for bitfield
      // access.
      BinaryOperator *Shift = dyn_cast<BinaryOperator>(LHSI->getOperand(0));
      if (Shift && !Shift->isShift())
        Shift = nullptr;

      ConstantInt *ShAmt;
      ShAmt = Shift ? dyn_cast<ConstantInt>(Shift->getOperand(1)) : nullptr;

      // This seemingly simple opportunity to fold away a shift turns out to
      // be rather complicated. See PR17827
      // ( http://llvm.org/bugs/show_bug.cgi?id=17827 ) for details.
      if (ShAmt) {
        bool CanFold = false;
        unsigned ShiftOpcode = Shift->getOpcode();
        if (ShiftOpcode == Instruction::AShr) {
          // There may be some constraints that make this possible,
          // but nothing simple has been discovered yet.
          CanFold = false;
        } else if (ShiftOpcode == Instruction::Shl) {
          // For a left shift, we can fold if the comparison is not signed.
          // We can also fold a signed comparison if the mask value and
          // comparison value are not negative. These constraints may not be
          // obvious, but we can prove that they are correct using an SMT
          // solver.
          if (!ICI.isSigned() || (!AndCst->isNegative() && !RHS->isNegative()))
            CanFold = true;
        } else if (ShiftOpcode == Instruction::LShr) {
          // For a logical right shift, we can fold if the comparison is not
          // signed. We can also fold a signed comparison if the shifted mask
          // value and the shifted comparison value are not negative.
          // These constraints may not be obvious, but we can prove that they
          // are correct using an SMT solver.
          if (!ICI.isSigned())
            CanFold = true;
          else {
            ConstantInt *ShiftedAndCst =
              cast<ConstantInt>(ConstantExpr::getShl(AndCst, ShAmt));
            ConstantInt *ShiftedRHSCst =
              cast<ConstantInt>(ConstantExpr::getShl(RHS, ShAmt));
            
            if (!ShiftedAndCst->isNegative() && !ShiftedRHSCst->isNegative())
              CanFold = true;
          }
        }

        if (CanFold) {
          Constant *NewCst;
          if (ShiftOpcode == Instruction::Shl)
            NewCst = ConstantExpr::getLShr(RHS, ShAmt);
          else
            NewCst = ConstantExpr::getShl(RHS, ShAmt);

          // Check to see if we are shifting out any of the bits being
          // compared.
          if (ConstantExpr::get(ShiftOpcode, NewCst, ShAmt) != RHS) {
            // If we shifted bits out, the fold is not going to work out.
            // As a special case, check to see if this means that the
            // result is always true or false now.
            if (ICI.getPredicate() == ICmpInst::ICMP_EQ)
              return ReplaceInstUsesWith(ICI, Builder->getFalse());
            if (ICI.getPredicate() == ICmpInst::ICMP_NE)
              return ReplaceInstUsesWith(ICI, Builder->getTrue());
          } else {
            ICI.setOperand(1, NewCst);
            Constant *NewAndCst;
            if (ShiftOpcode == Instruction::Shl)
              NewAndCst = ConstantExpr::getLShr(AndCst, ShAmt);
            else
              NewAndCst = ConstantExpr::getShl(AndCst, ShAmt);
            LHSI->setOperand(1, NewAndCst);
            LHSI->setOperand(0, Shift->getOperand(0));
            Worklist.Add(Shift); // Shift is dead.
            return &ICI;
          }
        }
      }

      // Turn ((X >> Y) & C) == 0  into  (X & (C << Y)) == 0.  The later is
      // preferable because it allows the C<<Y expression to be hoisted out
      // of a loop if Y is invariant and X is not.
      if (Shift && Shift->hasOneUse() && RHSV == 0 &&
          ICI.isEquality() && !Shift->isArithmeticShift() &&
          !isa<Constant>(Shift->getOperand(0))) {
        // Compute C << Y.
        Value *NS;
        if (Shift->getOpcode() == Instruction::LShr) {
          NS = Builder->CreateShl(AndCst, Shift->getOperand(1));
        } else {
          // Insert a logical shift.
          NS = Builder->CreateLShr(AndCst, Shift->getOperand(1));
        }

        // Compute X & (C << Y).
        Value *NewAnd =
          Builder->CreateAnd(Shift->getOperand(0), NS, LHSI->getName());

        ICI.setOperand(0, NewAnd);
        return &ICI;
      }

      // Replace ((X & AndCst) > RHSV) with ((X & AndCst) != 0), if any
      // bit set in (X & AndCst) will produce a result greater than RHSV.
      if (ICI.getPredicate() == ICmpInst::ICMP_UGT) {
        unsigned NTZ = AndCst->getValue().countTrailingZeros();
        if ((NTZ < AndCst->getBitWidth()) &&
            APInt::getOneBitSet(AndCst->getBitWidth(), NTZ).ugt(RHSV))
          return new ICmpInst(ICmpInst::ICMP_NE, LHSI,
                              Constant::getNullValue(RHS->getType()));
      }
    }

    // Try to optimize things like "A[i]&42 == 0" to index computations.
    if (LoadInst *LI = dyn_cast<LoadInst>(LHSI->getOperand(0))) {
      if (GetElementPtrInst *GEP =
          dyn_cast<GetElementPtrInst>(LI->getOperand(0)))
        if (GlobalVariable *GV = dyn_cast<GlobalVariable>(GEP->getOperand(0)))
          if (GV->isConstant() && GV->hasDefinitiveInitializer() &&
              !LI->isVolatile() && isa<ConstantInt>(LHSI->getOperand(1))) {
            ConstantInt *C = cast<ConstantInt>(LHSI->getOperand(1));
            if (Instruction *Res = FoldCmpLoadFromIndexedGlobal(GEP, GV,ICI, C))
              return Res;
          }
    }

    // X & -C == -C -> X >  u ~C
    // X & -C != -C -> X <= u ~C
    //   iff C is a power of 2
    if (ICI.isEquality() && RHS == LHSI->getOperand(1) && (-RHSV).isPowerOf2())
      return new ICmpInst(
          ICI.getPredicate() == ICmpInst::ICMP_EQ ? ICmpInst::ICMP_UGT
                                                  : ICmpInst::ICMP_ULE,
          LHSI->getOperand(0), SubOne(RHS));
    break;

  case Instruction::Or: {
    if (!ICI.isEquality() || !RHS->isNullValue() || !LHSI->hasOneUse())
      break;
    Value *P, *Q;
    if (match(LHSI, m_Or(m_PtrToInt(m_Value(P)), m_PtrToInt(m_Value(Q))))) {
      // Simplify icmp eq (or (ptrtoint P), (ptrtoint Q)), 0
      // -> and (icmp eq P, null), (icmp eq Q, null).
      Value *ICIP = Builder->CreateICmp(ICI.getPredicate(), P,
                                        Constant::getNullValue(P->getType()));
      Value *ICIQ = Builder->CreateICmp(ICI.getPredicate(), Q,
                                        Constant::getNullValue(Q->getType()));
      Instruction *Op;
      if (ICI.getPredicate() == ICmpInst::ICMP_EQ)
        Op = BinaryOperator::CreateAnd(ICIP, ICIQ);
      else
        Op = BinaryOperator::CreateOr(ICIP, ICIQ);
      return Op;
    }
    break;
  }

  case Instruction::Mul: {       // (icmp pred (mul X, Val), CI)
    ConstantInt *Val = dyn_cast<ConstantInt>(LHSI->getOperand(1));
    if (!Val) break;

    // If this is a signed comparison to 0 and the mul is sign preserving,
    // use the mul LHS operand instead.
    ICmpInst::Predicate pred = ICI.getPredicate();
    if (isSignTest(pred, RHS) && !Val->isZero() &&
        cast<BinaryOperator>(LHSI)->hasNoSignedWrap())
      return new ICmpInst(Val->isNegative() ?
                          ICmpInst::getSwappedPredicate(pred) : pred,
                          LHSI->getOperand(0),
                          Constant::getNullValue(RHS->getType()));

    break;
  }

  case Instruction::Shl: {       // (icmp pred (shl X, ShAmt), CI)
    uint32_t TypeBits = RHSV.getBitWidth();
    ConstantInt *ShAmt = dyn_cast<ConstantInt>(LHSI->getOperand(1));
    if (!ShAmt) {
      Value *X;
      // (1 << X) pred P2 -> X pred Log2(P2)
      if (match(LHSI, m_Shl(m_One(), m_Value(X)))) {
        bool RHSVIsPowerOf2 = RHSV.isPowerOf2();
        ICmpInst::Predicate Pred = ICI.getPredicate();
        if (ICI.isUnsigned()) {
          if (!RHSVIsPowerOf2) {
            // (1 << X) <  30 -> X <= 4
            // (1 << X) <= 30 -> X <= 4
            // (1 << X) >= 30 -> X >  4
            // (1 << X) >  30 -> X >  4
            if (Pred == ICmpInst::ICMP_ULT)
              Pred = ICmpInst::ICMP_ULE;
            else if (Pred == ICmpInst::ICMP_UGE)
              Pred = ICmpInst::ICMP_UGT;
          }
          unsigned RHSLog2 = RHSV.logBase2();

          // (1 << X) >= 2147483648 -> X >= 31 -> X == 31
          // (1 << X) >  2147483648 -> X >  31 -> false
          // (1 << X) <= 2147483648 -> X <= 31 -> true
          // (1 << X) <  2147483648 -> X <  31 -> X != 31
          if (RHSLog2 == TypeBits-1) {
            if (Pred == ICmpInst::ICMP_UGE)
              Pred = ICmpInst::ICMP_EQ;
            else if (Pred == ICmpInst::ICMP_UGT)
              return ReplaceInstUsesWith(ICI, Builder->getFalse());
            else if (Pred == ICmpInst::ICMP_ULE)
              return ReplaceInstUsesWith(ICI, Builder->getTrue());
            else if (Pred == ICmpInst::ICMP_ULT)
              Pred = ICmpInst::ICMP_NE;
          }

          return new ICmpInst(Pred, X,
                              ConstantInt::get(RHS->getType(), RHSLog2));
        } else if (ICI.isSigned()) {
          if (RHSV.isAllOnesValue()) {
            // (1 << X) <= -1 -> X == 31
            if (Pred == ICmpInst::ICMP_SLE)
              return new ICmpInst(ICmpInst::ICMP_EQ, X,
                                  ConstantInt::get(RHS->getType(), TypeBits-1));

            // (1 << X) >  -1 -> X != 31
            if (Pred == ICmpInst::ICMP_SGT)
              return new ICmpInst(ICmpInst::ICMP_NE, X,
                                  ConstantInt::get(RHS->getType(), TypeBits-1));
          } else if (!RHSV) {
            // (1 << X) <  0 -> X == 31
            // (1 << X) <= 0 -> X == 31
            if (Pred == ICmpInst::ICMP_SLT || Pred == ICmpInst::ICMP_SLE)
              return new ICmpInst(ICmpInst::ICMP_EQ, X,
                                  ConstantInt::get(RHS->getType(), TypeBits-1));

            // (1 << X) >= 0 -> X != 31
            // (1 << X) >  0 -> X != 31
            if (Pred == ICmpInst::ICMP_SGT || Pred == ICmpInst::ICMP_SGE)
              return new ICmpInst(ICmpInst::ICMP_NE, X,
                                  ConstantInt::get(RHS->getType(), TypeBits-1));
          }
        } else if (ICI.isEquality()) {
          if (RHSVIsPowerOf2)
            return new ICmpInst(
                Pred, X, ConstantInt::get(RHS->getType(), RHSV.logBase2()));

          return ReplaceInstUsesWith(
              ICI, Pred == ICmpInst::ICMP_EQ ? Builder->getFalse()
                                             : Builder->getTrue());
        }
      }
      break;
    }

    // Check that the shift amount is in range.  If not, don't perform
    // undefined shifts.  When the shift is visited it will be
    // simplified.
    if (ShAmt->uge(TypeBits))
      break;

    if (ICI.isEquality()) {
      // If we are comparing against bits always shifted out, the
      // comparison cannot succeed.
      Constant *Comp =
        ConstantExpr::getShl(ConstantExpr::getLShr(RHS, ShAmt),
                                                                 ShAmt);
      if (Comp != RHS) {// Comparing against a bit that we know is zero.
        bool IsICMP_NE = ICI.getPredicate() == ICmpInst::ICMP_NE;
        Constant *Cst = Builder->getInt1(IsICMP_NE);
        return ReplaceInstUsesWith(ICI, Cst);
      }

      // If the shift is NUW, then it is just shifting out zeros, no need for an
      // AND.
      if (cast<BinaryOperator>(LHSI)->hasNoUnsignedWrap())
        return new ICmpInst(ICI.getPredicate(), LHSI->getOperand(0),
                            ConstantExpr::getLShr(RHS, ShAmt));

      // If the shift is NSW and we compare to 0, then it is just shifting out
      // sign bits, no need for an AND either.
      if (cast<BinaryOperator>(LHSI)->hasNoSignedWrap() && RHSV == 0)
        return new ICmpInst(ICI.getPredicate(), LHSI->getOperand(0),
                            ConstantExpr::getLShr(RHS, ShAmt));

      if (LHSI->hasOneUse()) {
        // Otherwise strength reduce the shift into an and.
        uint32_t ShAmtVal = (uint32_t)ShAmt->getLimitedValue(TypeBits);
        Constant *Mask = Builder->getInt(APInt::getLowBitsSet(TypeBits,
                                                          TypeBits - ShAmtVal));

        Value *And =
          Builder->CreateAnd(LHSI->getOperand(0),Mask, LHSI->getName()+".mask");
        return new ICmpInst(ICI.getPredicate(), And,
                            ConstantExpr::getLShr(RHS, ShAmt));
      }
    }

    // If this is a signed comparison to 0 and the shift is sign preserving,
    // use the shift LHS operand instead.
    ICmpInst::Predicate pred = ICI.getPredicate();
    if (isSignTest(pred, RHS) &&
        cast<BinaryOperator>(LHSI)->hasNoSignedWrap())
      return new ICmpInst(pred,
                          LHSI->getOperand(0),
                          Constant::getNullValue(RHS->getType()));

    // Otherwise, if this is a comparison of the sign bit, simplify to and/test.
    bool TrueIfSigned = false;
    if (LHSI->hasOneUse() &&
        isSignBitCheck(ICI.getPredicate(), RHS, TrueIfSigned)) {
      // (X << 31) <s 0  --> (X&1) != 0
      Constant *Mask = ConstantInt::get(LHSI->getOperand(0)->getType(),
                                        APInt::getOneBitSet(TypeBits,
                                            TypeBits-ShAmt->getZExtValue()-1));
      Value *And =
        Builder->CreateAnd(LHSI->getOperand(0), Mask, LHSI->getName()+".mask");
      return new ICmpInst(TrueIfSigned ? ICmpInst::ICMP_NE : ICmpInst::ICMP_EQ,
                          And, Constant::getNullValue(And->getType()));
    }

    // Transform (icmp pred iM (shl iM %v, N), CI)
    // -> (icmp pred i(M-N) (trunc %v iM to i(M-N)), (trunc (CI>>N))
    // Transform the shl to a trunc if (trunc (CI>>N)) has no loss and M-N.
    // This enables to get rid of the shift in favor of a trunc which can be
    // free on the target. It has the additional benefit of comparing to a
    // smaller constant, which will be target friendly.
    unsigned Amt = ShAmt->getLimitedValue(TypeBits-1);
    if (LHSI->hasOneUse() &&
        Amt != 0 && RHSV.countTrailingZeros() >= Amt) {
      Type *NTy = IntegerType::get(ICI.getContext(), TypeBits - Amt);
      Constant *NCI = ConstantExpr::getTrunc(
                        ConstantExpr::getAShr(RHS,
                          ConstantInt::get(RHS->getType(), Amt)),
                        NTy);
      return new ICmpInst(ICI.getPredicate(),
                          Builder->CreateTrunc(LHSI->getOperand(0), NTy),
                          NCI);
    }

    break;
  }

  case Instruction::LShr:         // (icmp pred (shr X, ShAmt), CI)
  case Instruction::AShr: {
    // Handle equality comparisons of shift-by-constant.
    BinaryOperator *BO = cast<BinaryOperator>(LHSI);
    if (ConstantInt *ShAmt = dyn_cast<ConstantInt>(LHSI->getOperand(1))) {
      if (Instruction *Res = FoldICmpShrCst(ICI, BO, ShAmt))
        return Res;
    }

    // Handle exact shr's.
    if (ICI.isEquality() && BO->isExact() && BO->hasOneUse()) {
      if (RHSV.isMinValue())
        return new ICmpInst(ICI.getPredicate(), BO->getOperand(0), RHS);
    }
    break;
  }

  case Instruction::SDiv:
  case Instruction::UDiv:
    // Fold: icmp pred ([us]div X, C1), C2 -> range test
    // Fold this div into the comparison, producing a range check.
    // Determine, based on the divide type, what the range is being
    // checked.  If there is an overflow on the low or high side, remember
    // it, otherwise compute the range [low, hi) bounding the new value.
    // See: InsertRangeTest above for the kinds of replacements possible.
    if (ConstantInt *DivRHS = dyn_cast<ConstantInt>(LHSI->getOperand(1)))
      if (Instruction *R = FoldICmpDivCst(ICI, cast<BinaryOperator>(LHSI),
                                          DivRHS))
        return R;
    break;

  case Instruction::Sub: {
    ConstantInt *LHSC = dyn_cast<ConstantInt>(LHSI->getOperand(0));
    if (!LHSC) break;
    const APInt &LHSV = LHSC->getValue();

    // C1-X <u C2 -> (X|(C2-1)) == C1
    //   iff C1 & (C2-1) == C2-1
    //       C2 is a power of 2
    if (ICI.getPredicate() == ICmpInst::ICMP_ULT && LHSI->hasOneUse() &&
        RHSV.isPowerOf2() && (LHSV & (RHSV - 1)) == (RHSV - 1))
      return new ICmpInst(ICmpInst::ICMP_EQ,
                          Builder->CreateOr(LHSI->getOperand(1), RHSV - 1),
                          LHSC);

    // C1-X >u C2 -> (X|C2) != C1
    //   iff C1 & C2 == C2
    //       C2+1 is a power of 2
    if (ICI.getPredicate() == ICmpInst::ICMP_UGT && LHSI->hasOneUse() &&
        (RHSV + 1).isPowerOf2() && (LHSV & RHSV) == RHSV)
      return new ICmpInst(ICmpInst::ICMP_NE,
                          Builder->CreateOr(LHSI->getOperand(1), RHSV), LHSC);
    break;
  }

  case Instruction::Add:
    // Fold: icmp pred (add X, C1), C2
    if (!ICI.isEquality()) {
      ConstantInt *LHSC = dyn_cast<ConstantInt>(LHSI->getOperand(1));
      if (!LHSC) break;
      const APInt &LHSV = LHSC->getValue();

      ConstantRange CR = ICI.makeConstantRange(ICI.getPredicate(), RHSV)
                            .subtract(LHSV);

      if (ICI.isSigned()) {
        if (CR.getLower().isSignBit()) {
          return new ICmpInst(ICmpInst::ICMP_SLT, LHSI->getOperand(0),
                              Builder->getInt(CR.getUpper()));
        } else if (CR.getUpper().isSignBit()) {
          return new ICmpInst(ICmpInst::ICMP_SGE, LHSI->getOperand(0),
                              Builder->getInt(CR.getLower()));
        }
      } else {
        if (CR.getLower().isMinValue()) {
          return new ICmpInst(ICmpInst::ICMP_ULT, LHSI->getOperand(0),
                              Builder->getInt(CR.getUpper()));
        } else if (CR.getUpper().isMinValue()) {
          return new ICmpInst(ICmpInst::ICMP_UGE, LHSI->getOperand(0),
                              Builder->getInt(CR.getLower()));
        }
      }

      // X-C1 <u C2 -> (X & -C2) == C1
      //   iff C1 & (C2-1) == 0
      //       C2 is a power of 2
      if (ICI.getPredicate() == ICmpInst::ICMP_ULT && LHSI->hasOneUse() &&
          RHSV.isPowerOf2() && (LHSV & (RHSV - 1)) == 0)
        return new ICmpInst(ICmpInst::ICMP_EQ,
                            Builder->CreateAnd(LHSI->getOperand(0), -RHSV),
                            ConstantExpr::getNeg(LHSC));

      // X-C1 >u C2 -> (X & ~C2) != C1
      //   iff C1 & C2 == 0
      //       C2+1 is a power of 2
      if (ICI.getPredicate() == ICmpInst::ICMP_UGT && LHSI->hasOneUse() &&
          (RHSV + 1).isPowerOf2() && (LHSV & RHSV) == 0)
        return new ICmpInst(ICmpInst::ICMP_NE,
                            Builder->CreateAnd(LHSI->getOperand(0), ~RHSV),
                            ConstantExpr::getNeg(LHSC));
    }
    break;
  }

  // Simplify icmp_eq and icmp_ne instructions with integer constant RHS.
  if (ICI.isEquality()) {
    bool isICMP_NE = ICI.getPredicate() == ICmpInst::ICMP_NE;

    // If the first operand is (add|sub|and|or|xor|rem) with a constant, and
    // the second operand is a constant, simplify a bit.
    if (BinaryOperator *BO = dyn_cast<BinaryOperator>(LHSI)) {
      switch (BO->getOpcode()) {
      case Instruction::SRem:
        // If we have a signed (X % (2^c)) == 0, turn it into an unsigned one.
        if (RHSV == 0 && isa<ConstantInt>(BO->getOperand(1)) &&BO->hasOneUse()){
          const APInt &V = cast<ConstantInt>(BO->getOperand(1))->getValue();
          if (V.sgt(1) && V.isPowerOf2()) {
            Value *NewRem =
              Builder->CreateURem(BO->getOperand(0), BO->getOperand(1),
                                  BO->getName());
            return new ICmpInst(ICI.getPredicate(), NewRem,
                                Constant::getNullValue(BO->getType()));
          }
        }
        break;
      case Instruction::Add:
        // Replace ((add A, B) != C) with (A != C-B) if B & C are constants.
        if (ConstantInt *BOp1C = dyn_cast<ConstantInt>(BO->getOperand(1))) {
          if (BO->hasOneUse())
            return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
                                ConstantExpr::getSub(RHS, BOp1C));
        } else if (RHSV == 0) {
          // Replace ((add A, B) != 0) with (A != -B) if A or B is
          // efficiently invertible, or if the add has just this one use.
          Value *BOp0 = BO->getOperand(0), *BOp1 = BO->getOperand(1);

          if (Value *NegVal = dyn_castNegVal(BOp1))
            return new ICmpInst(ICI.getPredicate(), BOp0, NegVal);
          if (Value *NegVal = dyn_castNegVal(BOp0))
            return new ICmpInst(ICI.getPredicate(), NegVal, BOp1);
          if (BO->hasOneUse()) {
            Value *Neg = Builder->CreateNeg(BOp1);
            Neg->takeName(BO);
            return new ICmpInst(ICI.getPredicate(), BOp0, Neg);
          }
        }
        break;
      case Instruction::Xor:
        // For the xor case, we can xor two constants together, eliminating
        // the explicit xor.
        if (Constant *BOC = dyn_cast<Constant>(BO->getOperand(1))) {
          return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
                              ConstantExpr::getXor(RHS, BOC));
        } else if (RHSV == 0) {
          // Replace ((xor A, B) != 0) with (A != B)
          return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
                              BO->getOperand(1));
        }
        break;
      case Instruction::Sub:
        // Replace ((sub A, B) != C) with (B != A-C) if A & C are constants.
        if (ConstantInt *BOp0C = dyn_cast<ConstantInt>(BO->getOperand(0))) {
          if (BO->hasOneUse())
            return new ICmpInst(ICI.getPredicate(), BO->getOperand(1),
                                ConstantExpr::getSub(BOp0C, RHS));
        } else if (RHSV == 0) {
          // Replace ((sub A, B) != 0) with (A != B)
          return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
                              BO->getOperand(1));
        }
        break;
      case Instruction::Or:
        // If bits are being or'd in that are not present in the constant we
        // are comparing against, then the comparison could never succeed!
        if (ConstantInt *BOC = dyn_cast<ConstantInt>(BO->getOperand(1))) {
          Constant *NotCI = ConstantExpr::getNot(RHS);
          if (!ConstantExpr::getAnd(BOC, NotCI)->isNullValue())
            return ReplaceInstUsesWith(ICI, Builder->getInt1(isICMP_NE));
        }
        break;

      case Instruction::And:
        if (ConstantInt *BOC = dyn_cast<ConstantInt>(BO->getOperand(1))) {
          // If bits are being compared against that are and'd out, then the
          // comparison can never succeed!
          if ((RHSV & ~BOC->getValue()) != 0)
            return ReplaceInstUsesWith(ICI, Builder->getInt1(isICMP_NE));

          // If we have ((X & C) == C), turn it into ((X & C) != 0).
          if (RHS == BOC && RHSV.isPowerOf2())
            return new ICmpInst(isICMP_NE ? ICmpInst::ICMP_EQ :
                                ICmpInst::ICMP_NE, LHSI,
                                Constant::getNullValue(RHS->getType()));

          // Don't perform the following transforms if the AND has multiple uses
          if (!BO->hasOneUse())
            break;

          // Replace (and X, (1 << size(X)-1) != 0) with x s< 0
          if (BOC->getValue().isSignBit()) {
            Value *X = BO->getOperand(0);
            Constant *Zero = Constant::getNullValue(X->getType());
            ICmpInst::Predicate pred = isICMP_NE ?
              ICmpInst::ICMP_SLT : ICmpInst::ICMP_SGE;
            return new ICmpInst(pred, X, Zero);
          }

          // ((X & ~7) == 0) --> X < 8
          if (RHSV == 0 && isHighOnes(BOC)) {
            Value *X = BO->getOperand(0);
            Constant *NegX = ConstantExpr::getNeg(BOC);
            ICmpInst::Predicate pred = isICMP_NE ?
              ICmpInst::ICMP_UGE : ICmpInst::ICMP_ULT;
            return new ICmpInst(pred, X, NegX);
          }
        }
        break;
      case Instruction::Mul:
        if (RHSV == 0 && BO->hasNoSignedWrap()) {
          if (ConstantInt *BOC = dyn_cast<ConstantInt>(BO->getOperand(1))) {
            // The trivial case (mul X, 0) is handled by InstSimplify
            // General case : (mul X, C) != 0 iff X != 0
            //                (mul X, C) == 0 iff X == 0
            if (!BOC->isZero())
              return new ICmpInst(ICI.getPredicate(), BO->getOperand(0),
                                  Constant::getNullValue(RHS->getType()));
          }
        }
        break;
      default: break;
      }
    } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(LHSI)) {
      // Handle icmp {eq|ne} <intrinsic>, intcst.
      switch (II->getIntrinsicID()) {
      case Intrinsic::bswap:
        Worklist.Add(II);
        ICI.setOperand(0, II->getArgOperand(0));
        ICI.setOperand(1, Builder->getInt(RHSV.byteSwap()));
        return &ICI;
      case Intrinsic::ctlz:
      case Intrinsic::cttz:
        // ctz(A) == bitwidth(a)  ->  A == 0 and likewise for !=
        if (RHSV == RHS->getType()->getBitWidth()) {
          Worklist.Add(II);
          ICI.setOperand(0, II->getArgOperand(0));
          ICI.setOperand(1, ConstantInt::get(RHS->getType(), 0));
          return &ICI;
        }
        break;
      case Intrinsic::ctpop:
        // popcount(A) == 0  ->  A == 0 and likewise for !=
        if (RHS->isZero()) {
          Worklist.Add(II);
          ICI.setOperand(0, II->getArgOperand(0));
          ICI.setOperand(1, RHS);
          return &ICI;
        }
        break;
      default:
        break;
      }
    }
  }
  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitIndirectBrInst ( IndirectBrInst &  I )

inlineinherited

Definition at line 169 of file InstVisitor.h.

{ DELEGATE(TerminatorInst);}

Instruction * InstCombiner::visitInsertElementInst

(

InsertElementInst &

IE

)

Definition at line 528 of file InstCombineVectorOps.cpp.

                                                                       {
  Value *VecOp    = IE.getOperand(0);
  Value *ScalarOp = IE.getOperand(1);
  Value *IdxOp    = IE.getOperand(2);

  // Inserting an undef or into an undefined place, remove this.
  if (isa<UndefValue>(ScalarOp) || isa<UndefValue>(IdxOp))
    ReplaceInstUsesWith(IE, VecOp);

  // If the inserted element was extracted from some other vector, and if the
  // indexes are constant, try to turn this into a shufflevector operation.
  if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)) {
    if (isa<ConstantInt>(EI->getOperand(1)) && isa<ConstantInt>(IdxOp)) {
      unsigned NumInsertVectorElts = IE.getType()->getNumElements();
      unsigned NumExtractVectorElts =
          EI->getOperand(0)->getType()->getVectorNumElements();
      unsigned ExtractedIdx =
        cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
      unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();

      if (ExtractedIdx >= NumExtractVectorElts) // Out of range extract.
        return ReplaceInstUsesWith(IE, VecOp);

      if (InsertedIdx >= NumInsertVectorElts)  // Out of range insert.
        return ReplaceInstUsesWith(IE, UndefValue::get(IE.getType()));

      // If we are extracting a value from a vector, then inserting it right
      // back into the same place, just use the input vector.
      if (EI->getOperand(0) == VecOp && ExtractedIdx == InsertedIdx)
        return ReplaceInstUsesWith(IE, VecOp);

      // If this insertelement isn't used by some other insertelement, turn it
      // (and any insertelements it points to), into one big shuffle.
      if (!IE.hasOneUse() || !isa<InsertElementInst>(IE.user_back())) {
        SmallVector<Constant*, 16> Mask;
        ShuffleOps LR = CollectShuffleElements(&IE, Mask, nullptr);

        // The proposed shuffle may be trivial, in which case we shouldn't
        // perform the combine.
        if (LR.first != &IE && LR.second != &IE) {
          // We now have a shuffle of LHS, RHS, Mask.
          if (LR.second == nullptr)
            LR.second = UndefValue::get(LR.first->getType());
          return new ShuffleVectorInst(LR.first, LR.second,
                                       ConstantVector::get(Mask));
        }
      }
    }
  }

  unsigned VWidth = cast<VectorType>(VecOp->getType())->getNumElements();
  APInt UndefElts(VWidth, 0);
  APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
  if (Value *V = SimplifyDemandedVectorElts(&IE, AllOnesEltMask, UndefElts)) {
    if (V != &IE)
      return ReplaceInstUsesWith(IE, V);
    return &IE;
  }

  return nullptr;
}

Instruction * InstCombiner::visitInsertValueInst

(

InsertValueInst &

I

)

Try to find redundant insertvalue instructions, like the following ones: %0 = insertvalue { i8, i32 } undef, i8 x, 0 %1 = insertvalue { i8, i32 } %0, i8 y, 0 Here the second instruction inserts values at the same indices, as the first one, making the first one redundant. It should be transformed to: %0 = insertvalue { i8, i32 } undef, i8 y, 0

Definition at line 500 of file InstCombineVectorOps.cpp.

                                                                  {
  bool IsRedundant = false;
  ArrayRef<unsigned int> FirstIndices = I.getIndices();

  // If there is a chain of insertvalue instructions (each of them except the
  // last one has only one use and it's another insertvalue insn from this
  // chain), check if any of the 'children' uses the same indices as the first
  // instruction. In this case, the first one is redundant.
  Value *V = &I;
  unsigned Depth = 0;
  while (V->hasOneUse() && Depth < 10) {
    User *U = V->user_back();
    auto UserInsInst = dyn_cast<InsertValueInst>(U);
    if (!UserInsInst || U->getOperand(0) != V)
      break;
    if (UserInsInst->getIndices() == FirstIndices) {
      IsRedundant = true;
      break;
    }
    V = UserInsInst;
    Depth++;
  }

  if (IsRedundant)
    return ReplaceInstUsesWith(I, I.getOperand(0));
  return nullptr;
}

Instruction* llvm::InstCombiner::visitInstruction ( Instruction &  I )

inline

Definition at line 224 of file InstCombine.h.

{ return nullptr; }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitIntrinsicInst ( IntrinsicInst &  I )

inlineinherited

Definition at line 216 of file InstVisitor.h.

{ DELEGATE(CallInst); }

Instruction * InstCombiner::visitIntToPtr

(

IntToPtrInst &

CI

)

Definition at line 1406 of file InstCombineCasts.cpp.

                                                         {
  // If the source integer type is not the intptr_t type for this target, do a
  // trunc or zext to the intptr_t type, then inttoptr of it.  This allows the
  // cast to be exposed to other transforms.

  if (DL) {
    unsigned AS = CI.getAddressSpace();
    if (CI.getOperand(0)->getType()->getScalarSizeInBits() !=
        DL->getPointerSizeInBits(AS)) {
      Type *Ty = DL->getIntPtrType(CI.getContext(), AS);
      if (CI.getType()->isVectorTy()) // Handle vectors of pointers.
        Ty = VectorType::get(Ty, CI.getType()->getVectorNumElements());

      Value *P = Builder->CreateZExtOrTrunc(CI.getOperand(0), Ty);
      return new IntToPtrInst(P, CI.getType());
    }
  }

  if (Instruction *I = commonCastTransforms(CI))
    return I;

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitIntToPtrInst ( IntToPtrInst &  I )

inlineinherited

Definition at line 192 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitInvokeInst

(

InvokeInst &

II

)

Definition at line 1004 of file InstCombineCalls.cpp.

                                                         {
  return visitCallSite(&II);
}

Instruction * InstCombiner::visitLandingPadInst

(

LandingPadInst &

LI

)

Definition at line 2193 of file InstructionCombining.cpp.

                                                                 {
  // The logic here should be correct for any real-world personality function.
  // However if that turns out not to be true, the offending logic can always
  // be conditioned on the personality function, like the catch-all logic is.
  Personality_Type Personality = RecognizePersonality(LI.getPersonalityFn());

  // Simplify the list of clauses, eg by removing repeated catch clauses
  // (these are often created by inlining).
  bool MakeNewInstruction = false; // If true, recreate using the following:
  SmallVector<Constant *, 16> NewClauses; // - Clauses for the new instruction;
  bool CleanupFlag = LI.isCleanup();   // - The new instruction is a cleanup.

  SmallPtrSet<Value *, 16> AlreadyCaught; // Typeinfos known caught already.
  for (unsigned i = 0, e = LI.getNumClauses(); i != e; ++i) {
    bool isLastClause = i + 1 == e;
    if (LI.isCatch(i)) {
      // A catch clause.
      Constant *CatchClause = LI.getClause(i);
      Constant *TypeInfo = CatchClause->stripPointerCasts();

      // If we already saw this clause, there is no point in having a second
      // copy of it.
      if (AlreadyCaught.insert(TypeInfo)) {
        // This catch clause was not already seen.
        NewClauses.push_back(CatchClause);
      } else {
        // Repeated catch clause - drop the redundant copy.
        MakeNewInstruction = true;
      }

      // If this is a catch-all then there is no point in keeping any following
      // clauses or marking the landingpad as having a cleanup.
      if (isCatchAll(Personality, TypeInfo)) {
        if (!isLastClause)
          MakeNewInstruction = true;
        CleanupFlag = false;
        break;
      }
    } else {
      // A filter clause.  If any of the filter elements were already caught
      // then they can be dropped from the filter.  It is tempting to try to
      // exploit the filter further by saying that any typeinfo that does not
      // occur in the filter can't be caught later (and thus can be dropped).
      // However this would be wrong, since typeinfos can match without being
      // equal (for example if one represents a C++ class, and the other some
      // class derived from it).
      assert(LI.isFilter(i) && "Unsupported landingpad clause!");
      Constant *FilterClause = LI.getClause(i);
      ArrayType *FilterType = cast<ArrayType>(FilterClause->getType());
      unsigned NumTypeInfos = FilterType->getNumElements();

      // An empty filter catches everything, so there is no point in keeping any
      // following clauses or marking the landingpad as having a cleanup.  By
      // dealing with this case here the following code is made a bit simpler.
      if (!NumTypeInfos) {
        NewClauses.push_back(FilterClause);
        if (!isLastClause)
          MakeNewInstruction = true;
        CleanupFlag = false;
        break;
      }

      bool MakeNewFilter = false; // If true, make a new filter.
      SmallVector<Constant *, 16> NewFilterElts; // New elements.
      if (isa<ConstantAggregateZero>(FilterClause)) {
        // Not an empty filter - it contains at least one null typeinfo.
        assert(NumTypeInfos > 0 && "Should have handled empty filter already!");
        Constant *TypeInfo =
          Constant::getNullValue(FilterType->getElementType());
        // If this typeinfo is a catch-all then the filter can never match.
        if (isCatchAll(Personality, TypeInfo)) {
          // Throw the filter away.
          MakeNewInstruction = true;
          continue;
        }

        // There is no point in having multiple copies of this typeinfo, so
        // discard all but the first copy if there is more than one.
        NewFilterElts.push_back(TypeInfo);
        if (NumTypeInfos > 1)
          MakeNewFilter = true;
      } else {
        ConstantArray *Filter = cast<ConstantArray>(FilterClause);
        SmallPtrSet<Value *, 16> SeenInFilter; // For uniquing the elements.
        NewFilterElts.reserve(NumTypeInfos);

        // Remove any filter elements that were already caught or that already
        // occurred in the filter.  While there, see if any of the elements are
        // catch-alls.  If so, the filter can be discarded.
        bool SawCatchAll = false;
        for (unsigned j = 0; j != NumTypeInfos; ++j) {
          Constant *Elt = Filter->getOperand(j);
          Constant *TypeInfo = Elt->stripPointerCasts();
          if (isCatchAll(Personality, TypeInfo)) {
            // This element is a catch-all.  Bail out, noting this fact.
            SawCatchAll = true;
            break;
          }
          if (AlreadyCaught.count(TypeInfo))
            // Already caught by an earlier clause, so having it in the filter
            // is pointless.
            continue;
          // There is no point in having multiple copies of the same typeinfo in
          // a filter, so only add it if we didn't already.
          if (SeenInFilter.insert(TypeInfo))
            NewFilterElts.push_back(cast<Constant>(Elt));
        }
        // A filter containing a catch-all cannot match anything by definition.
        if (SawCatchAll) {
          // Throw the filter away.
          MakeNewInstruction = true;
          continue;
        }

        // If we dropped something from the filter, make a new one.
        if (NewFilterElts.size() < NumTypeInfos)
          MakeNewFilter = true;
      }
      if (MakeNewFilter) {
        FilterType = ArrayType::get(FilterType->getElementType(),
                                    NewFilterElts.size());
        FilterClause = ConstantArray::get(FilterType, NewFilterElts);
        MakeNewInstruction = true;
      }

      NewClauses.push_back(FilterClause);

      // If the new filter is empty then it will catch everything so there is
      // no point in keeping any following clauses or marking the landingpad
      // as having a cleanup.  The case of the original filter being empty was
      // already handled above.
      if (MakeNewFilter && !NewFilterElts.size()) {
        assert(MakeNewInstruction && "New filter but not a new instruction!");
        CleanupFlag = false;
        break;
      }
    }
  }

  // If several filters occur in a row then reorder them so that the shortest
  // filters come first (those with the smallest number of elements).  This is
  // advantageous because shorter filters are more likely to match, speeding up
  // unwinding, but mostly because it increases the effectiveness of the other
  // filter optimizations below.
  for (unsigned i = 0, e = NewClauses.size(); i + 1 < e; ) {
    unsigned j;
    // Find the maximal 'j' s.t. the range [i, j) consists entirely of filters.
    for (j = i; j != e; ++j)
      if (!isa<ArrayType>(NewClauses[j]->getType()))
        break;

    // Check whether the filters are already sorted by length.  We need to know
    // if sorting them is actually going to do anything so that we only make a
    // new landingpad instruction if it does.
    for (unsigned k = i; k + 1 < j; ++k)
      if (shorter_filter(NewClauses[k+1], NewClauses[k])) {
        // Not sorted, so sort the filters now.  Doing an unstable sort would be
        // correct too but reordering filters pointlessly might confuse users.
        std::stable_sort(NewClauses.begin() + i, NewClauses.begin() + j,
                         shorter_filter);
        MakeNewInstruction = true;
        break;
      }

    // Look for the next batch of filters.
    i = j + 1;
  }

  // If typeinfos matched if and only if equal, then the elements of a filter L
  // that occurs later than a filter F could be replaced by the intersection of
  // the elements of F and L.  In reality two typeinfos can match without being
  // equal (for example if one represents a C++ class, and the other some class
  // derived from it) so it would be wrong to perform this transform in general.
  // However the transform is correct and useful if F is a subset of L.  In that
  // case L can be replaced by F, and thus removed altogether since repeating a
  // filter is pointless.  So here we look at all pairs of filters F and L where
  // L follows F in the list of clauses, and remove L if every element of F is
  // an element of L.  This can occur when inlining C++ functions with exception
  // specifications.
  for (unsigned i = 0; i + 1 < NewClauses.size(); ++i) {
    // Examine each filter in turn.
    Value *Filter = NewClauses[i];
    ArrayType *FTy = dyn_cast<ArrayType>(Filter->getType());
    if (!FTy)
      // Not a filter - skip it.
      continue;
    unsigned FElts = FTy->getNumElements();
    // Examine each filter following this one.  Doing this backwards means that
    // we don't have to worry about filters disappearing under us when removed.
    for (unsigned j = NewClauses.size() - 1; j != i; --j) {
      Value *LFilter = NewClauses[j];
      ArrayType *LTy = dyn_cast<ArrayType>(LFilter->getType());
      if (!LTy)
        // Not a filter - skip it.
        continue;
      // If Filter is a subset of LFilter, i.e. every element of Filter is also
      // an element of LFilter, then discard LFilter.
      SmallVectorImpl<Constant *>::iterator J = NewClauses.begin() + j;
      // If Filter is empty then it is a subset of LFilter.
      if (!FElts) {
        // Discard LFilter.
        NewClauses.erase(J);
        MakeNewInstruction = true;
        // Move on to the next filter.
        continue;
      }
      unsigned LElts = LTy->getNumElements();
      // If Filter is longer than LFilter then it cannot be a subset of it.
      if (FElts > LElts)
        // Move on to the next filter.
        continue;
      // At this point we know that LFilter has at least one element.
      if (isa<ConstantAggregateZero>(LFilter)) { // LFilter only contains zeros.
        // Filter is a subset of LFilter iff Filter contains only zeros (as we
        // already know that Filter is not longer than LFilter).
        if (isa<ConstantAggregateZero>(Filter)) {
          assert(FElts <= LElts && "Should have handled this case earlier!");
          // Discard LFilter.
          NewClauses.erase(J);
          MakeNewInstruction = true;
        }
        // Move on to the next filter.
        continue;
      }
      ConstantArray *LArray = cast<ConstantArray>(LFilter);
      if (isa<ConstantAggregateZero>(Filter)) { // Filter only contains zeros.
        // Since Filter is non-empty and contains only zeros, it is a subset of
        // LFilter iff LFilter contains a zero.
        assert(FElts > 0 && "Should have eliminated the empty filter earlier!");
        for (unsigned l = 0; l != LElts; ++l)
          if (LArray->getOperand(l)->isNullValue()) {
            // LFilter contains a zero - discard it.
            NewClauses.erase(J);
            MakeNewInstruction = true;
            break;
          }
        // Move on to the next filter.
        continue;
      }
      // At this point we know that both filters are ConstantArrays.  Loop over
      // operands to see whether every element of Filter is also an element of
      // LFilter.  Since filters tend to be short this is probably faster than
      // using a method that scales nicely.
      ConstantArray *FArray = cast<ConstantArray>(Filter);
      bool AllFound = true;
      for (unsigned f = 0; f != FElts; ++f) {
        Value *FTypeInfo = FArray->getOperand(f)->stripPointerCasts();
        AllFound = false;
        for (unsigned l = 0; l != LElts; ++l) {
          Value *LTypeInfo = LArray->getOperand(l)->stripPointerCasts();
          if (LTypeInfo == FTypeInfo) {
            AllFound = true;
            break;
          }
        }
        if (!AllFound)
          break;
      }
      if (AllFound) {
        // Discard LFilter.
        NewClauses.erase(J);
        MakeNewInstruction = true;
      }
      // Move on to the next filter.
    }
  }

  // If we changed any of the clauses, replace the old landingpad instruction
  // with a new one.
  if (MakeNewInstruction) {
    LandingPadInst *NLI = LandingPadInst::Create(LI.getType(),
                                                 LI.getPersonalityFn(),
                                                 NewClauses.size());
    for (unsigned i = 0, e = NewClauses.size(); i != e; ++i)
      NLI->addClause(NewClauses[i]);
    // A landing pad with no clauses must have the cleanup flag set.  It is
    // theoretically possible, though highly unlikely, that we eliminated all
    // clauses.  If so, force the cleanup flag to true.
    if (NewClauses.empty())
      CleanupFlag = true;
    NLI->setCleanup(CleanupFlag);
    return NLI;
  }

  // Even if none of the clauses changed, we may nonetheless have understood
  // that the cleanup flag is pointless.  Clear it if so.
  if (LI.isCleanup() != CleanupFlag) {
    assert(!CleanupFlag && "Adding a cleanup, not removing one?!");
    LI.setCleanup(CleanupFlag);
    return &LI;
  }

  return nullptr;
}

Instruction * InstCombiner::visitLoadInst

(

LoadInst &

LI

)

Definition at line 360 of file InstCombineLoadStoreAlloca.cpp.

                                                     {
  Value *Op = LI.getOperand(0);

  // Attempt to improve the alignment.
  if (DL) {
    unsigned KnownAlign =
      getOrEnforceKnownAlignment(Op, DL->getPrefTypeAlignment(LI.getType()),DL);
    unsigned LoadAlign = LI.getAlignment();
    unsigned EffectiveLoadAlign = LoadAlign != 0 ? LoadAlign :
      DL->getABITypeAlignment(LI.getType());

    if (KnownAlign > EffectiveLoadAlign)
      LI.setAlignment(KnownAlign);
    else if (LoadAlign == 0)
      LI.setAlignment(EffectiveLoadAlign);
  }

  // load (cast X) --> cast (load X) iff safe.
  if (isa<CastInst>(Op))
    if (Instruction *Res = InstCombineLoadCast(*this, LI, DL))
      return Res;

  // None of the following transforms are legal for volatile/atomic loads.
  // FIXME: Some of it is okay for atomic loads; needs refactoring.
  if (!LI.isSimple()) return nullptr;

  // Do really simple store-to-load forwarding and load CSE, to catch cases
  // where there are several consecutive memory accesses to the same location,
  // separated by a few arithmetic operations.
  BasicBlock::iterator BBI = &LI;
  if (Value *AvailableVal = FindAvailableLoadedValue(Op, LI.getParent(), BBI,6))
    return ReplaceInstUsesWith(LI, AvailableVal);

  // load(gep null, ...) -> unreachable
  if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(Op)) {
    const Value *GEPI0 = GEPI->getOperand(0);
    // TODO: Consider a target hook for valid address spaces for this xform.
    if (isa<ConstantPointerNull>(GEPI0) && GEPI->getPointerAddressSpace() == 0){
      // Insert a new store to null instruction before the load to indicate
      // that this code is not reachable.  We do this instead of inserting
      // an unreachable instruction directly because we cannot modify the
      // CFG.
      new StoreInst(UndefValue::get(LI.getType()),
                    Constant::getNullValue(Op->getType()), &LI);
      return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
    }
  }

  // load null/undef -> unreachable
  // TODO: Consider a target hook for valid address spaces for this xform.
  if (isa<UndefValue>(Op) ||
      (isa<ConstantPointerNull>(Op) && LI.getPointerAddressSpace() == 0)) {
    // Insert a new store to null instruction before the load to indicate that
    // this code is not reachable.  We do this instead of inserting an
    // unreachable instruction directly because we cannot modify the CFG.
    new StoreInst(UndefValue::get(LI.getType()),
                  Constant::getNullValue(Op->getType()), &LI);
    return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
  }

  // Instcombine load (constantexpr_cast global) -> cast (load global)
  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Op))
    if (CE->isCast())
      if (Instruction *Res = InstCombineLoadCast(*this, LI, DL))
        return Res;

  if (Op->hasOneUse()) {
    // Change select and PHI nodes to select values instead of addresses: this
    // helps alias analysis out a lot, allows many others simplifications, and
    // exposes redundancy in the code.
    //
    // Note that we cannot do the transformation unless we know that the
    // introduced loads cannot trap!  Something like this is valid as long as
    // the condition is always false: load (select bool %C, int* null, int* %G),
    // but it would not be valid if we transformed it to load from null
    // unconditionally.
    //
    if (SelectInst *SI = dyn_cast<SelectInst>(Op)) {
      // load (select (Cond, &V1, &V2))  --> select(Cond, load &V1, load &V2).
      unsigned Align = LI.getAlignment();
      if (isSafeToLoadUnconditionally(SI->getOperand(1), SI, Align, DL) &&
          isSafeToLoadUnconditionally(SI->getOperand(2), SI, Align, DL)) {
        LoadInst *V1 = Builder->CreateLoad(SI->getOperand(1),
                                           SI->getOperand(1)->getName()+".val");
        LoadInst *V2 = Builder->CreateLoad(SI->getOperand(2),
                                           SI->getOperand(2)->getName()+".val");
        V1->setAlignment(Align);
        V2->setAlignment(Align);
        return SelectInst::Create(SI->getCondition(), V1, V2);
      }

      // load (select (cond, null, P)) -> load P
      if (Constant *C = dyn_cast<Constant>(SI->getOperand(1)))
        if (C->isNullValue()) {
          LI.setOperand(0, SI->getOperand(2));
          return &LI;
        }

      // load (select (cond, P, null)) -> load P
      if (Constant *C = dyn_cast<Constant>(SI->getOperand(2)))
        if (C->isNullValue()) {
          LI.setOperand(0, SI->getOperand(1));
          return &LI;
        }
    }
  }
  return nullptr;
}

Instruction * InstCombiner::visitLShr

(

BinaryOperator &

I

)

Definition at line 729 of file InstCombineShifts.cpp.

                                                      {
  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyLShrInst(I.getOperand(0), I.getOperand(1),
                                  I.isExact(), DL))
    return ReplaceInstUsesWith(I, V);

  if (Instruction *R = commonShiftTransforms(I))
    return R;

  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (ConstantInt *Op1C = dyn_cast<ConstantInt>(Op1)) {
    unsigned ShAmt = Op1C->getZExtValue();

    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Op0)) {
      unsigned BitWidth = Op0->getType()->getScalarSizeInBits();
      // ctlz.i32(x)>>5  --> zext(x == 0)
      // cttz.i32(x)>>5  --> zext(x == 0)
      // ctpop.i32(x)>>5 --> zext(x == -1)
      if ((II->getIntrinsicID() == Intrinsic::ctlz ||
           II->getIntrinsicID() == Intrinsic::cttz ||
           II->getIntrinsicID() == Intrinsic::ctpop) &&
          isPowerOf2_32(BitWidth) && Log2_32(BitWidth) == ShAmt) {
        bool isCtPop = II->getIntrinsicID() == Intrinsic::ctpop;
        Constant *RHS = ConstantInt::getSigned(Op0->getType(), isCtPop ? -1:0);
        Value *Cmp = Builder->CreateICmpEQ(II->getArgOperand(0), RHS);
        return new ZExtInst(Cmp, II->getType());
      }
    }

    // If the shifted-out value is known-zero, then this is an exact shift.
    if (!I.isExact() &&
        MaskedValueIsZero(Op0,APInt::getLowBitsSet(Op1C->getBitWidth(),ShAmt))){
      I.setIsExact();
      return &I;
    }
  }

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitMemCpyInst ( MemCpyInst &  I )

inlineinherited

Definition at line 209 of file InstVisitor.h.

{ DELEGATE(MemTransferInst); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitMemIntrinsic ( MemIntrinsic &  I )

inlineinherited

Definition at line 212 of file InstVisitor.h.

{ DELEGATE(IntrinsicInst); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitMemMoveInst ( MemMoveInst &  I )

inlineinherited

Definition at line 210 of file InstVisitor.h.

{ DELEGATE(MemTransferInst); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitMemSetInst ( MemSetInst &  I )

inlineinherited

Definition at line 208 of file InstVisitor.h.

{ DELEGATE(MemIntrinsic); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitMemTransferInst ( MemTransferInst &  I )

inlineinherited

Definition at line 211 of file InstVisitor.h.

{ DELEGATE(MemIntrinsic); }

void llvm::InstVisitor< InstCombiner , Instruction * >::visitModule ( Module &  M )

inlineinherited

Definition at line 141 of file InstVisitor.h.

{}

Instruction * InstCombiner::visitMul

(

BinaryOperator &

I

)

i1 mul -> i1 and.

Definition at line 119 of file InstCombineMulDivRem.cpp.

                                                     {
  bool Changed = SimplifyAssociativeOrCommutative(I);
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyMulInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyUsingDistributiveLaws(I))
    return ReplaceInstUsesWith(I, V);

  if (match(Op1, m_AllOnes()))  // X * -1 == 0 - X
    return BinaryOperator::CreateNeg(Op0, I.getName());

  // Also allow combining multiply instructions on vectors.
  {
    Value *NewOp;
    Constant *C1, *C2;
    const APInt *IVal;
    if (match(&I, m_Mul(m_Shl(m_Value(NewOp), m_Constant(C2)),
                        m_Constant(C1))) &&
        match(C1, m_APInt(IVal)))
      // ((X << C1)*C2) == (X * (C2 << C1))
      return BinaryOperator::CreateMul(NewOp, ConstantExpr::getShl(C1, C2));

    if (match(&I, m_Mul(m_Value(NewOp), m_Constant(C1)))) {
      Constant *NewCst = nullptr;
      if (match(C1, m_APInt(IVal)) && IVal->isPowerOf2())
        // Replace X*(2^C) with X << C, where C is either a scalar or a splat.
        NewCst = ConstantInt::get(NewOp->getType(), IVal->logBase2());
      else if (ConstantDataVector *CV = dyn_cast<ConstantDataVector>(C1))
        // Replace X*(2^C) with X << C, where C is a vector of known
        // constant powers of 2.
        NewCst = getLogBase2Vector(CV);

      if (NewCst) {
        BinaryOperator *Shl = BinaryOperator::CreateShl(NewOp, NewCst);
        if (I.hasNoSignedWrap()) Shl->setHasNoSignedWrap();
        if (I.hasNoUnsignedWrap()) Shl->setHasNoUnsignedWrap();
        return Shl;
      }
    }
  }

  if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
    // (Y - X) * (-(2**n)) -> (X - Y) * (2**n), for positive nonzero n
    // (Y + const) * (-(2**n)) -> (-constY) * (2**n), for positive nonzero n
    // The "* (2**n)" thus becomes a potential shifting opportunity.
    {
      const APInt &   Val = CI->getValue();
      const APInt &PosVal = Val.abs();
      if (Val.isNegative() && PosVal.isPowerOf2()) {
        Value *X = nullptr, *Y = nullptr;
        if (Op0->hasOneUse()) {
          ConstantInt *C1;
          Value *Sub = nullptr;
          if (match(Op0, m_Sub(m_Value(Y), m_Value(X))))
            Sub = Builder->CreateSub(X, Y, "suba");
          else if (match(Op0, m_Add(m_Value(Y), m_ConstantInt(C1))))
            Sub = Builder->CreateSub(Builder->CreateNeg(C1), Y, "subc");
          if (Sub)
            return
              BinaryOperator::CreateMul(Sub,
                                        ConstantInt::get(Y->getType(), PosVal));
        }
      }
    }
  }

  // Simplify mul instructions with a constant RHS.
  if (isa<Constant>(Op1)) {
    // Try to fold constant mul into select arguments.
    if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;

    if (isa<PHINode>(Op0))
      if (Instruction *NV = FoldOpIntoPhi(I))
        return NV;

    // Canonicalize (X+C1)*CI -> X*CI+C1*CI.
    {
      Value *X;
      Constant *C1;
      if (match(Op0, m_OneUse(m_Add(m_Value(X), m_Constant(C1))))) {
        Value *Mul = Builder->CreateMul(C1, Op1);
        // Only go forward with the transform if C1*CI simplifies to a tidier
        // constant.
        if (!match(Mul, m_Mul(m_Value(), m_Value())))
          return BinaryOperator::CreateAdd(Builder->CreateMul(X, Op1), Mul);
      }
    }
  }

  if (Value *Op0v = dyn_castNegVal(Op0))     // -X * -Y = X*Y
    if (Value *Op1v = dyn_castNegVal(Op1))
      return BinaryOperator::CreateMul(Op0v, Op1v);

  // (X / Y) *  Y = X - (X % Y)
  // (X / Y) * -Y = (X % Y) - X
  {
    Value *Op1C = Op1;
    BinaryOperator *BO = dyn_cast<BinaryOperator>(Op0);
    if (!BO ||
        (BO->getOpcode() != Instruction::UDiv &&
         BO->getOpcode() != Instruction::SDiv)) {
      Op1C = Op0;
      BO = dyn_cast<BinaryOperator>(Op1);
    }
    Value *Neg = dyn_castNegVal(Op1C);
    if (BO && BO->hasOneUse() &&
        (BO->getOperand(1) == Op1C || BO->getOperand(1) == Neg) &&
        (BO->getOpcode() == Instruction::UDiv ||
         BO->getOpcode() == Instruction::SDiv)) {
      Value *Op0BO = BO->getOperand(0), *Op1BO = BO->getOperand(1);

      // If the division is exact, X % Y is zero, so we end up with X or -X.
      if (PossiblyExactOperator *SDiv = dyn_cast<PossiblyExactOperator>(BO))
        if (SDiv->isExact()) {
          if (Op1BO == Op1C)
            return ReplaceInstUsesWith(I, Op0BO);
          return BinaryOperator::CreateNeg(Op0BO);
        }

      Value *Rem;
      if (BO->getOpcode() == Instruction::UDiv)
        Rem = Builder->CreateURem(Op0BO, Op1BO);
      else
        Rem = Builder->CreateSRem(Op0BO, Op1BO);
      Rem->takeName(BO);

      if (Op1BO == Op1C)
        return BinaryOperator::CreateSub(Op0BO, Rem);
      return BinaryOperator::CreateSub(Rem, Op0BO);
    }
  }

  /// i1 mul -> i1 and.
  if (I.getType()->getScalarType()->isIntegerTy(1))
    return BinaryOperator::CreateAnd(Op0, Op1);

  // X*(1 << Y) --> X << Y
  // (1 << Y)*X --> X << Y
  {
    Value *Y;
    if (match(Op0, m_Shl(m_One(), m_Value(Y))))
      return BinaryOperator::CreateShl(Op1, Y);
    if (match(Op1, m_Shl(m_One(), m_Value(Y))))
      return BinaryOperator::CreateShl(Op0, Y);
  }

  // If one of the operands of the multiply is a cast from a boolean value, then
  // we know the bool is either zero or one, so this is a 'masking' multiply.
  //   X * Y (where Y is 0 or 1) -> X & (0-Y)
  if (!I.getType()->isVectorTy()) {
    // -2 is "-1 << 1" so it is all bits set except the low one.
    APInt Negative2(I.getType()->getPrimitiveSizeInBits(), (uint64_t)-2, true);

    Value *BoolCast = nullptr, *OtherOp = nullptr;
    if (MaskedValueIsZero(Op0, Negative2))
      BoolCast = Op0, OtherOp = Op1;
    else if (MaskedValueIsZero(Op1, Negative2))
      BoolCast = Op1, OtherOp = Op0;

    if (BoolCast) {
      Value *V = Builder->CreateSub(Constant::getNullValue(I.getType()),
                                    BoolCast);
      return BinaryOperator::CreateAnd(V, OtherOp);
    }
  }

  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitOr

(

BinaryOperator &

I

)

Definition at line 1909 of file InstCombineAndOrXor.cpp.

                                                    {
  bool Changed = SimplifyAssociativeOrCommutative(I);
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyOrInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // (A&B)|(A&C) -> A&(B|C) etc
  if (Value *V = SimplifyUsingDistributiveLaws(I))
    return ReplaceInstUsesWith(I, V);

  // See if we can simplify any instructions used by the instruction whose sole
  // purpose is to compute bits we don't care about.
  if (SimplifyDemandedInstructionBits(I))
    return &I;

  if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
    ConstantInt *C1 = nullptr; Value *X = nullptr;
    // (X & C1) | C2 --> (X | C2) & (C1|C2)
    // iff (C1 & C2) == 0.
    if (match(Op0, m_And(m_Value(X), m_ConstantInt(C1))) &&
        (RHS->getValue() & C1->getValue()) != 0 &&
        Op0->hasOneUse()) {
      Value *Or = Builder->CreateOr(X, RHS);
      Or->takeName(Op0);
      return BinaryOperator::CreateAnd(Or,
                             Builder->getInt(RHS->getValue() | C1->getValue()));
    }

    // (X ^ C1) | C2 --> (X | C2) ^ (C1&~C2)
    if (match(Op0, m_Xor(m_Value(X), m_ConstantInt(C1))) &&
        Op0->hasOneUse()) {
      Value *Or = Builder->CreateOr(X, RHS);
      Or->takeName(Op0);
      return BinaryOperator::CreateXor(Or,
                            Builder->getInt(C1->getValue() & ~RHS->getValue()));
    }

    // Try to fold constant and into select arguments.
    if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;

    if (isa<PHINode>(Op0))
      if (Instruction *NV = FoldOpIntoPhi(I))
        return NV;
  }

  Value *A = nullptr, *B = nullptr;
  ConstantInt *C1 = nullptr, *C2 = nullptr;

  // (A | B) | C  and  A | (B | C)                  -> bswap if possible.
  // (A >> B) | (C << D)  and  (A << B) | (B >> C)  -> bswap if possible.
  if (match(Op0, m_Or(m_Value(), m_Value())) ||
      match(Op1, m_Or(m_Value(), m_Value())) ||
      (match(Op0, m_LogicalShift(m_Value(), m_Value())) &&
       match(Op1, m_LogicalShift(m_Value(), m_Value())))) {
    if (Instruction *BSwap = MatchBSwap(I))
      return BSwap;
  }

  // (X^C)|Y -> (X|Y)^C iff Y&C == 0
  if (Op0->hasOneUse() &&
      match(Op0, m_Xor(m_Value(A), m_ConstantInt(C1))) &&
      MaskedValueIsZero(Op1, C1->getValue())) {
    Value *NOr = Builder->CreateOr(A, Op1);
    NOr->takeName(Op0);
    return BinaryOperator::CreateXor(NOr, C1);
  }

  // Y|(X^C) -> (X|Y)^C iff Y&C == 0
  if (Op1->hasOneUse() &&
      match(Op1, m_Xor(m_Value(A), m_ConstantInt(C1))) &&
      MaskedValueIsZero(Op0, C1->getValue())) {
    Value *NOr = Builder->CreateOr(A, Op0);
    NOr->takeName(Op0);
    return BinaryOperator::CreateXor(NOr, C1);
  }

  // (A & C)|(B & D)
  Value *C = nullptr, *D = nullptr;
  if (match(Op0, m_And(m_Value(A), m_Value(C))) &&
      match(Op1, m_And(m_Value(B), m_Value(D)))) {
    Value *V1 = nullptr, *V2 = nullptr;
    C1 = dyn_cast<ConstantInt>(C);
    C2 = dyn_cast<ConstantInt>(D);
    if (C1 && C2) {  // (A & C1)|(B & C2)
      if ((C1->getValue() & C2->getValue()) == 0) {
        // ((V | N) & C1) | (V & C2) --> (V|N) & (C1|C2)
        // iff (C1&C2) == 0 and (N&~C1) == 0
        if (match(A, m_Or(m_Value(V1), m_Value(V2))) &&
            ((V1 == B && MaskedValueIsZero(V2, ~C1->getValue())) ||  // (V|N)
             (V2 == B && MaskedValueIsZero(V1, ~C1->getValue()))))   // (N|V)
          return BinaryOperator::CreateAnd(A,
                                Builder->getInt(C1->getValue()|C2->getValue()));
        // Or commutes, try both ways.
        if (match(B, m_Or(m_Value(V1), m_Value(V2))) &&
            ((V1 == A && MaskedValueIsZero(V2, ~C2->getValue())) ||  // (V|N)
             (V2 == A && MaskedValueIsZero(V1, ~C2->getValue()))))   // (N|V)
          return BinaryOperator::CreateAnd(B,
                                Builder->getInt(C1->getValue()|C2->getValue()));

        // ((V|C3)&C1) | ((V|C4)&C2) --> (V|C3|C4)&(C1|C2)
        // iff (C1&C2) == 0 and (C3&~C1) == 0 and (C4&~C2) == 0.
        ConstantInt *C3 = nullptr, *C4 = nullptr;
        if (match(A, m_Or(m_Value(V1), m_ConstantInt(C3))) &&
            (C3->getValue() & ~C1->getValue()) == 0 &&
            match(B, m_Or(m_Specific(V1), m_ConstantInt(C4))) &&
            (C4->getValue() & ~C2->getValue()) == 0) {
          V2 = Builder->CreateOr(V1, ConstantExpr::getOr(C3, C4), "bitfield");
          return BinaryOperator::CreateAnd(V2,
                                Builder->getInt(C1->getValue()|C2->getValue()));
        }
      }
    }

    // (A & (C0?-1:0)) | (B & ~(C0?-1:0)) ->  C0 ? A : B, and commuted variants.
    // Don't do this for vector select idioms, the code generator doesn't handle
    // them well yet.
    if (!I.getType()->isVectorTy()) {
      if (Instruction *Match = MatchSelectFromAndOr(A, B, C, D))
        return Match;
      if (Instruction *Match = MatchSelectFromAndOr(B, A, D, C))
        return Match;
      if (Instruction *Match = MatchSelectFromAndOr(C, B, A, D))
        return Match;
      if (Instruction *Match = MatchSelectFromAndOr(D, A, B, C))
        return Match;
    }

    // ((A&~B)|(~A&B)) -> A^B
    if ((match(C, m_Not(m_Specific(D))) &&
         match(B, m_Not(m_Specific(A)))))
      return BinaryOperator::CreateXor(A, D);
    // ((~B&A)|(~A&B)) -> A^B
    if ((match(A, m_Not(m_Specific(D))) &&
         match(B, m_Not(m_Specific(C)))))
      return BinaryOperator::CreateXor(C, D);
    // ((A&~B)|(B&~A)) -> A^B
    if ((match(C, m_Not(m_Specific(B))) &&
         match(D, m_Not(m_Specific(A)))))
      return BinaryOperator::CreateXor(A, B);
    // ((~B&A)|(B&~A)) -> A^B
    if ((match(A, m_Not(m_Specific(B))) &&
         match(D, m_Not(m_Specific(C)))))
      return BinaryOperator::CreateXor(C, B);

    // ((A|B)&1)|(B&-2) -> (A&1) | B
    if (match(A, m_Or(m_Value(V1), m_Specific(B))) ||
        match(A, m_Or(m_Specific(B), m_Value(V1)))) {
      Instruction *Ret = FoldOrWithConstants(I, Op1, V1, B, C);
      if (Ret) return Ret;
    }
    // (B&-2)|((A|B)&1) -> (A&1) | B
    if (match(B, m_Or(m_Specific(A), m_Value(V1))) ||
        match(B, m_Or(m_Value(V1), m_Specific(A)))) {
      Instruction *Ret = FoldOrWithConstants(I, Op0, A, V1, D);
      if (Ret) return Ret;
    }
  }

  // (X >> Z) | (Y >> Z)  -> (X|Y) >> Z  for all shifts.
  if (BinaryOperator *SI1 = dyn_cast<BinaryOperator>(Op1)) {
    if (BinaryOperator *SI0 = dyn_cast<BinaryOperator>(Op0))
      if (SI0->isShift() && SI0->getOpcode() == SI1->getOpcode() &&
          SI0->getOperand(1) == SI1->getOperand(1) &&
          (SI0->hasOneUse() || SI1->hasOneUse())) {
        Value *NewOp = Builder->CreateOr(SI0->getOperand(0), SI1->getOperand(0),
                                         SI0->getName());
        return BinaryOperator::Create(SI1->getOpcode(), NewOp,
                                      SI1->getOperand(1));
      }
  }

  // (~A | ~B) == (~(A & B)) - De Morgan's Law
  if (Value *Op0NotVal = dyn_castNotVal(Op0))
    if (Value *Op1NotVal = dyn_castNotVal(Op1))
      if (Op0->hasOneUse() && Op1->hasOneUse()) {
        Value *And = Builder->CreateAnd(Op0NotVal, Op1NotVal,
                                        I.getName()+".demorgan");
        return BinaryOperator::CreateNot(And);
      }

  // Canonicalize xor to the RHS.
  bool SwappedForXor = false;
  if (match(Op0, m_Xor(m_Value(), m_Value()))) {
    std::swap(Op0, Op1);
    SwappedForXor = true;
  }

  // A | ( A ^ B) -> A |  B
  // A | (~A ^ B) -> A | ~B
  // (A & B) | (A ^ B)
  if (match(Op1, m_Xor(m_Value(A), m_Value(B)))) {
    if (Op0 == A || Op0 == B)
      return BinaryOperator::CreateOr(A, B);

    if (match(Op0, m_And(m_Specific(A), m_Specific(B))) ||
        match(Op0, m_And(m_Specific(B), m_Specific(A))))
      return BinaryOperator::CreateOr(A, B);

    if (Op1->hasOneUse() && match(A, m_Not(m_Specific(Op0)))) {
      Value *Not = Builder->CreateNot(B, B->getName()+".not");
      return BinaryOperator::CreateOr(Not, Op0);
    }
    if (Op1->hasOneUse() && match(B, m_Not(m_Specific(Op0)))) {
      Value *Not = Builder->CreateNot(A, A->getName()+".not");
      return BinaryOperator::CreateOr(Not, Op0);
    }
  }

  // A | ~(A | B) -> A | ~B
  // A | ~(A ^ B) -> A | ~B
  if (match(Op1, m_Not(m_Value(A))))
    if (BinaryOperator *B = dyn_cast<BinaryOperator>(A))
      if ((Op0 == B->getOperand(0) || Op0 == B->getOperand(1)) &&
          Op1->hasOneUse() && (B->getOpcode() == Instruction::Or ||
                               B->getOpcode() == Instruction::Xor)) {
        Value *NotOp = Op0 == B->getOperand(0) ? B->getOperand(1) :
                                                 B->getOperand(0);
        Value *Not = Builder->CreateNot(NotOp, NotOp->getName()+".not");
        return BinaryOperator::CreateOr(Not, Op0);
      }

  if (SwappedForXor)
    std::swap(Op0, Op1);

  if (ICmpInst *RHS = dyn_cast<ICmpInst>(I.getOperand(1)))
    if (ICmpInst *LHS = dyn_cast<ICmpInst>(I.getOperand(0)))
      if (Value *Res = FoldOrOfICmps(LHS, RHS))
        return ReplaceInstUsesWith(I, Res);

  // (fcmp uno x, c) | (fcmp uno y, c)  -> (fcmp uno x, y)
  if (FCmpInst *LHS = dyn_cast<FCmpInst>(I.getOperand(0)))
    if (FCmpInst *RHS = dyn_cast<FCmpInst>(I.getOperand(1)))
      if (Value *Res = FoldOrOfFCmps(LHS, RHS))
        return ReplaceInstUsesWith(I, Res);

  // fold (or (cast A), (cast B)) -> (cast (or A, B))
  if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
    CastInst *Op1C = dyn_cast<CastInst>(Op1);
    if (Op1C && Op0C->getOpcode() == Op1C->getOpcode()) {// same cast kind ?
      Type *SrcTy = Op0C->getOperand(0)->getType();
      if (SrcTy == Op1C->getOperand(0)->getType() &&
          SrcTy->isIntOrIntVectorTy()) {
        Value *Op0COp = Op0C->getOperand(0), *Op1COp = Op1C->getOperand(0);

        if ((!isa<ICmpInst>(Op0COp) || !isa<ICmpInst>(Op1COp)) &&
            // Only do this if the casts both really cause code to be
            // generated.
            ShouldOptimizeCast(Op0C->getOpcode(), Op0COp, I.getType()) &&
            ShouldOptimizeCast(Op1C->getOpcode(), Op1COp, I.getType())) {
          Value *NewOp = Builder->CreateOr(Op0COp, Op1COp, I.getName());
          return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
        }

        // If this is or(cast(icmp), cast(icmp)), try to fold this even if the
        // cast is otherwise not optimizable.  This happens for vector sexts.
        if (ICmpInst *RHS = dyn_cast<ICmpInst>(Op1COp))
          if (ICmpInst *LHS = dyn_cast<ICmpInst>(Op0COp))
            if (Value *Res = FoldOrOfICmps(LHS, RHS))
              return CastInst::Create(Op0C->getOpcode(), Res, I.getType());

        // If this is or(cast(fcmp), cast(fcmp)), try to fold this even if the
        // cast is otherwise not optimizable.  This happens for vector sexts.
        if (FCmpInst *RHS = dyn_cast<FCmpInst>(Op1COp))
          if (FCmpInst *LHS = dyn_cast<FCmpInst>(Op0COp))
            if (Value *Res = FoldOrOfFCmps(LHS, RHS))
              return CastInst::Create(Op0C->getOpcode(), Res, I.getType());
      }
    }
  }

  // or(sext(A), B) -> A ? -1 : B where A is an i1
  // or(A, sext(B)) -> B ? -1 : A where B is an i1
  if (match(Op0, m_SExt(m_Value(A))) && A->getType()->isIntegerTy(1))
    return SelectInst::Create(A, ConstantInt::getSigned(I.getType(), -1), Op1);
  if (match(Op1, m_SExt(m_Value(A))) && A->getType()->isIntegerTy(1))
    return SelectInst::Create(A, ConstantInt::getSigned(I.getType(), -1), Op0);

  // Note: If we've gotten to the point of visiting the outer OR, then the
  // inner one couldn't be simplified.  If it was a constant, then it won't
  // be simplified by a later pass either, so we try swapping the inner/outer
  // ORs in the hopes that we'll be able to simplify it this way.
  // (X|C) | V --> (X|V) | C
  if (Op0->hasOneUse() && !isa<ConstantInt>(Op1) &&
      match(Op0, m_Or(m_Value(A), m_ConstantInt(C1)))) {
    Value *Inner = Builder->CreateOr(A, Op1);
    Inner->takeName(Op0);
    return BinaryOperator::CreateOr(Inner, C1);
  }

  // Change (or (bool?A:B),(bool?C:D)) --> (bool?(or A,C):(or B,D))
  // Since this OR statement hasn't been optimized further yet, we hope
  // that this transformation will allow the new ORs to be optimized.
  {
    Value *X = nullptr, *Y = nullptr;
    if (Op0->hasOneUse() && Op1->hasOneUse() &&
        match(Op0, m_Select(m_Value(X), m_Value(A), m_Value(B))) &&
        match(Op1, m_Select(m_Value(Y), m_Value(C), m_Value(D))) && X == Y) {
      Value *orTrue = Builder->CreateOr(A, C);
      Value *orFalse = Builder->CreateOr(B, D);
      return SelectInst::Create(X, orTrue, orFalse);
    }
  }

  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitPHINode

(

PHINode &

PN

)

Definition at line 790 of file InstCombinePHI.cpp.

                                                   {
  if (Value *V = SimplifyInstruction(&PN, DL, TLI))
    return ReplaceInstUsesWith(PN, V);

  // If all PHI operands are the same operation, pull them through the PHI,
  // reducing code size.
  if (isa<Instruction>(PN.getIncomingValue(0)) &&
      isa<Instruction>(PN.getIncomingValue(1)) &&
      cast<Instruction>(PN.getIncomingValue(0))->getOpcode() ==
      cast<Instruction>(PN.getIncomingValue(1))->getOpcode() &&
      // FIXME: The hasOneUse check will fail for PHIs that use the value more
      // than themselves more than once.
      PN.getIncomingValue(0)->hasOneUse())
    if (Instruction *Result = FoldPHIArgOpIntoPHI(PN))
      return Result;

  // If this is a trivial cycle in the PHI node graph, remove it.  Basically, if
  // this PHI only has a single use (a PHI), and if that PHI only has one use (a
  // PHI)... break the cycle.
  if (PN.hasOneUse()) {
    Instruction *PHIUser = cast<Instruction>(PN.user_back());
    if (PHINode *PU = dyn_cast<PHINode>(PHIUser)) {
      SmallPtrSet<PHINode*, 16> PotentiallyDeadPHIs;
      PotentiallyDeadPHIs.insert(&PN);
      if (DeadPHICycle(PU, PotentiallyDeadPHIs))
        return ReplaceInstUsesWith(PN, UndefValue::get(PN.getType()));
    }

    // If this phi has a single use, and if that use just computes a value for
    // the next iteration of a loop, delete the phi.  This occurs with unused
    // induction variables, e.g. "for (int j = 0; ; ++j);".  Detecting this
    // common case here is good because the only other things that catch this
    // are induction variable analysis (sometimes) and ADCE, which is only run
    // late.
    if (PHIUser->hasOneUse() &&
        (isa<BinaryOperator>(PHIUser) || isa<GetElementPtrInst>(PHIUser)) &&
        PHIUser->user_back() == &PN) {
      return ReplaceInstUsesWith(PN, UndefValue::get(PN.getType()));
    }
  }

  // We sometimes end up with phi cycles that non-obviously end up being the
  // same value, for example:
  //   z = some value; x = phi (y, z); y = phi (x, z)
  // where the phi nodes don't necessarily need to be in the same block.  Do a
  // quick check to see if the PHI node only contains a single non-phi value, if
  // so, scan to see if the phi cycle is actually equal to that value.
  {
    unsigned InValNo = 0, NumIncomingVals = PN.getNumIncomingValues();
    // Scan for the first non-phi operand.
    while (InValNo != NumIncomingVals &&
           isa<PHINode>(PN.getIncomingValue(InValNo)))
      ++InValNo;

    if (InValNo != NumIncomingVals) {
      Value *NonPhiInVal = PN.getIncomingValue(InValNo);

      // Scan the rest of the operands to see if there are any conflicts, if so
      // there is no need to recursively scan other phis.
      for (++InValNo; InValNo != NumIncomingVals; ++InValNo) {
        Value *OpVal = PN.getIncomingValue(InValNo);
        if (OpVal != NonPhiInVal && !isa<PHINode>(OpVal))
          break;
      }

      // If we scanned over all operands, then we have one unique value plus
      // phi values.  Scan PHI nodes to see if they all merge in each other or
      // the value.
      if (InValNo == NumIncomingVals) {
        SmallPtrSet<PHINode*, 16> ValueEqualPHIs;
        if (PHIsEqualValue(&PN, NonPhiInVal, ValueEqualPHIs))
          return ReplaceInstUsesWith(PN, NonPhiInVal);
      }
    }
  }

  // If there are multiple PHIs, sort their operands so that they all list
  // the blocks in the same order. This will help identical PHIs be eliminated
  // by other passes. Other passes shouldn't depend on this for correctness
  // however.
  PHINode *FirstPN = cast<PHINode>(PN.getParent()->begin());
  if (&PN != FirstPN)
    for (unsigned i = 0, e = FirstPN->getNumIncomingValues(); i != e; ++i) {
      BasicBlock *BBA = PN.getIncomingBlock(i);
      BasicBlock *BBB = FirstPN->getIncomingBlock(i);
      if (BBA != BBB) {
        Value *VA = PN.getIncomingValue(i);
        unsigned j = PN.getBasicBlockIndex(BBB);
        Value *VB = PN.getIncomingValue(j);
        PN.setIncomingBlock(i, BBB);
        PN.setIncomingValue(i, VB);
        PN.setIncomingBlock(j, BBA);
        PN.setIncomingValue(j, VA);
        // NOTE: Instcombine normally would want us to "return &PN" if we
        // modified any of the operands of an instruction.  However, since we
        // aren't adding or removing uses (just rearranging them) we don't do
        // this in this case.
      }
    }

  // If this is an integer PHI and we know that it has an illegal type, see if
  // it is only used by trunc or trunc(lshr) operations.  If so, we split the
  // PHI into the various pieces being extracted.  This sort of thing is
  // introduced when SROA promotes an aggregate to a single large integer type.
  if (PN.getType()->isIntegerTy() && DL &&
      !DL->isLegalInteger(PN.getType()->getPrimitiveSizeInBits()))
    if (Instruction *Res = SliceUpIllegalIntegerPHI(PN))
      return Res;

  return nullptr;
}

Instruction * InstCombiner::visitPtrToInt

(

PtrToIntInst &

CI

)

Definition at line 1490 of file InstCombineCasts.cpp.

                                                         {
  // If the destination integer type is not the intptr_t type for this target,
  // do a ptrtoint to intptr_t then do a trunc or zext.  This allows the cast
  // to be exposed to other transforms.

  if (!DL)
    return commonPointerCastTransforms(CI);

  Type *Ty = CI.getType();
  unsigned AS = CI.getPointerAddressSpace();

  if (Ty->getScalarSizeInBits() == DL->getPointerSizeInBits(AS))
    return commonPointerCastTransforms(CI);

  Type *PtrTy = DL->getIntPtrType(CI.getContext(), AS);
  if (Ty->isVectorTy()) // Handle vectors of pointers.
    PtrTy = VectorType::get(PtrTy, Ty->getVectorNumElements());

  Value *P = Builder->CreatePtrToInt(CI.getOperand(0), PtrTy);
  return CastInst::CreateIntegerCast(P, Ty, /*isSigned=*/false);
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitPtrToIntInst ( PtrToIntInst &  I )

inlineinherited

Definition at line 191 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitResumeInst ( ResumeInst &  I )

inlineinherited

Definition at line 170 of file InstVisitor.h.

{ DELEGATE(TerminatorInst);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitReturnInst ( ReturnInst &  I )

inlineinherited

Definition at line 166 of file InstVisitor.h.

{ DELEGATE(TerminatorInst);}

Instruction * InstCombiner::visitSDiv

(

BinaryOperator &

I

)

Definition at line 968 of file InstCombineMulDivRem.cpp.

                                                      {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifySDivInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // Handle the integer div common cases
  if (Instruction *Common = commonIDivTransforms(I))
    return Common;

  // sdiv X, -1 == -X
  if (match(Op1, m_AllOnes()))
    return BinaryOperator::CreateNeg(Op0);

  if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
    // sdiv X, C  -->  ashr exact X, log2(C)
    if (I.isExact() && RHS->getValue().isNonNegative() &&
        RHS->getValue().isPowerOf2()) {
      Value *ShAmt = llvm::ConstantInt::get(RHS->getType(),
                                            RHS->getValue().exactLogBase2());
      return BinaryOperator::CreateExactAShr(Op0, ShAmt, I.getName());
    }
  }

  if (Constant *RHS = dyn_cast<Constant>(Op1)) {
    // X/INT_MIN -> X == INT_MIN
    if (RHS->isMinSignedValue())
      return new ZExtInst(Builder->CreateICmpEQ(Op0, Op1), I.getType());

    // -X/C  -->  X/-C  provided the negation doesn't overflow.
    if (SubOperator *Sub = dyn_cast<SubOperator>(Op0))
      if (match(Sub->getOperand(0), m_Zero()) && Sub->hasNoSignedWrap())
        return BinaryOperator::CreateSDiv(Sub->getOperand(1),
                                          ConstantExpr::getNeg(RHS));
  }

  // If the sign bits of both operands are zero (i.e. we can prove they are
  // unsigned inputs), turn this into a udiv.
  if (I.getType()->isIntegerTy()) {
    APInt Mask(APInt::getSignBit(I.getType()->getPrimitiveSizeInBits()));
    if (MaskedValueIsZero(Op0, Mask)) {
      if (MaskedValueIsZero(Op1, Mask)) {
        // X sdiv Y -> X udiv Y, iff X and Y don't have sign bit set
        return BinaryOperator::CreateUDiv(Op0, Op1, I.getName());
      }

      if (match(Op1, m_Shl(m_Power2(), m_Value()))) {
        // X sdiv (1 << Y) -> X udiv (1 << Y) ( -> X u>> Y)
        // Safe because the only negative value (1 << Y) can take on is
        // INT_MIN, and X sdiv INT_MIN == X udiv INT_MIN == 0 if X doesn't have
        // the sign bit set.
        return BinaryOperator::CreateUDiv(Op0, Op1, I.getName());
      }
    }
  }

  return nullptr;
}

Instruction * InstCombiner::visitSelectInst

(

SelectInst &

SI

)

Definition at line 823 of file InstCombineSelect.cpp.

                                                         {
  Value *CondVal = SI.getCondition();
  Value *TrueVal = SI.getTrueValue();
  Value *FalseVal = SI.getFalseValue();

  if (Value *V = SimplifySelectInst(CondVal, TrueVal, FalseVal, DL))
    return ReplaceInstUsesWith(SI, V);

  if (SI.getType()->isIntegerTy(1)) {
    if (ConstantInt *C = dyn_cast<ConstantInt>(TrueVal)) {
      if (C->getZExtValue()) {
        // Change: A = select B, true, C --> A = or B, C
        return BinaryOperator::CreateOr(CondVal, FalseVal);
      }
      // Change: A = select B, false, C --> A = and !B, C
      Value *NotCond = Builder->CreateNot(CondVal, "not."+CondVal->getName());
      return BinaryOperator::CreateAnd(NotCond, FalseVal);
    }
    if (ConstantInt *C = dyn_cast<ConstantInt>(FalseVal)) {
      if (C->getZExtValue() == false) {
        // Change: A = select B, C, false --> A = and B, C
        return BinaryOperator::CreateAnd(CondVal, TrueVal);
      }
      // Change: A = select B, C, true --> A = or !B, C
      Value *NotCond = Builder->CreateNot(CondVal, "not."+CondVal->getName());
      return BinaryOperator::CreateOr(NotCond, TrueVal);
    }

    // select a, b, a  -> a&b
    // select a, a, b  -> a|b
    if (CondVal == TrueVal)
      return BinaryOperator::CreateOr(CondVal, FalseVal);
    if (CondVal == FalseVal)
      return BinaryOperator::CreateAnd(CondVal, TrueVal);

    // select a, ~a, b -> (~a)&b
    // select a, b, ~a -> (~a)|b
    if (match(TrueVal, m_Not(m_Specific(CondVal))))
      return BinaryOperator::CreateAnd(TrueVal, FalseVal);
    if (match(FalseVal, m_Not(m_Specific(CondVal))))
      return BinaryOperator::CreateOr(TrueVal, FalseVal);
  }

  // Selecting between two integer constants?
  if (ConstantInt *TrueValC = dyn_cast<ConstantInt>(TrueVal))
    if (ConstantInt *FalseValC = dyn_cast<ConstantInt>(FalseVal)) {
      // select C, 1, 0 -> zext C to int
      if (FalseValC->isZero() && TrueValC->getValue() == 1)
        return new ZExtInst(CondVal, SI.getType());

      // select C, -1, 0 -> sext C to int
      if (FalseValC->isZero() && TrueValC->isAllOnesValue())
        return new SExtInst(CondVal, SI.getType());

      // select C, 0, 1 -> zext !C to int
      if (TrueValC->isZero() && FalseValC->getValue() == 1) {
        Value *NotCond = Builder->CreateNot(CondVal, "not."+CondVal->getName());
        return new ZExtInst(NotCond, SI.getType());
      }

      // select C, 0, -1 -> sext !C to int
      if (TrueValC->isZero() && FalseValC->isAllOnesValue()) {
        Value *NotCond = Builder->CreateNot(CondVal, "not."+CondVal->getName());
        return new SExtInst(NotCond, SI.getType());
      }

      if (Value *V = foldSelectICmpAnd(SI, TrueValC, FalseValC, Builder))
        return ReplaceInstUsesWith(SI, V);
    }

  // See if we are selecting two values based on a comparison of the two values.
  if (FCmpInst *FCI = dyn_cast<FCmpInst>(CondVal)) {
    if (FCI->getOperand(0) == TrueVal && FCI->getOperand(1) == FalseVal) {
      // Transform (X == Y) ? X : Y  -> Y
      if (FCI->getPredicate() == FCmpInst::FCMP_OEQ) {
        // This is not safe in general for floating point:
        // consider X== -0, Y== +0.
        // It becomes safe if either operand is a nonzero constant.
        ConstantFP *CFPt, *CFPf;
        if (((CFPt = dyn_cast<ConstantFP>(TrueVal)) &&
              !CFPt->getValueAPF().isZero()) ||
            ((CFPf = dyn_cast<ConstantFP>(FalseVal)) &&
             !CFPf->getValueAPF().isZero()))
        return ReplaceInstUsesWith(SI, FalseVal);
      }
      // Transform (X une Y) ? X : Y  -> X
      if (FCI->getPredicate() == FCmpInst::FCMP_UNE) {
        // This is not safe in general for floating point:
        // consider X== -0, Y== +0.
        // It becomes safe if either operand is a nonzero constant.
        ConstantFP *CFPt, *CFPf;
        if (((CFPt = dyn_cast<ConstantFP>(TrueVal)) &&
              !CFPt->getValueAPF().isZero()) ||
            ((CFPf = dyn_cast<ConstantFP>(FalseVal)) &&
             !CFPf->getValueAPF().isZero()))
        return ReplaceInstUsesWith(SI, TrueVal);
      }
      // NOTE: if we wanted to, this is where to detect MIN/MAX

    } else if (FCI->getOperand(0) == FalseVal && FCI->getOperand(1) == TrueVal){
      // Transform (X == Y) ? Y : X  -> X
      if (FCI->getPredicate() == FCmpInst::FCMP_OEQ) {
        // This is not safe in general for floating point:
        // consider X== -0, Y== +0.
        // It becomes safe if either operand is a nonzero constant.
        ConstantFP *CFPt, *CFPf;
        if (((CFPt = dyn_cast<ConstantFP>(TrueVal)) &&
              !CFPt->getValueAPF().isZero()) ||
            ((CFPf = dyn_cast<ConstantFP>(FalseVal)) &&
             !CFPf->getValueAPF().isZero()))
          return ReplaceInstUsesWith(SI, FalseVal);
      }
      // Transform (X une Y) ? Y : X  -> Y
      if (FCI->getPredicate() == FCmpInst::FCMP_UNE) {
        // This is not safe in general for floating point:
        // consider X== -0, Y== +0.
        // It becomes safe if either operand is a nonzero constant.
        ConstantFP *CFPt, *CFPf;
        if (((CFPt = dyn_cast<ConstantFP>(TrueVal)) &&
              !CFPt->getValueAPF().isZero()) ||
            ((CFPf = dyn_cast<ConstantFP>(FalseVal)) &&
             !CFPf->getValueAPF().isZero()))
          return ReplaceInstUsesWith(SI, TrueVal);
      }
      // NOTE: if we wanted to, this is where to detect MIN/MAX
    }
    // NOTE: if we wanted to, this is where to detect ABS
  }

  // See if we are selecting two values based on a comparison of the two values.
  if (ICmpInst *ICI = dyn_cast<ICmpInst>(CondVal))
    if (Instruction *Result = visitSelectInstWithICmp(SI, ICI))
      return Result;

  if (Instruction *TI = dyn_cast<Instruction>(TrueVal))
    if (Instruction *FI = dyn_cast<Instruction>(FalseVal))
      if (TI->hasOneUse() && FI->hasOneUse()) {
        Instruction *AddOp = nullptr, *SubOp = nullptr;

        // Turn (select C, (op X, Y), (op X, Z)) -> (op X, (select C, Y, Z))
        if (TI->getOpcode() == FI->getOpcode())
          if (Instruction *IV = FoldSelectOpOp(SI, TI, FI))
            return IV;

        // Turn select C, (X+Y), (X-Y) --> (X+(select C, Y, (-Y))).  This is
        // even legal for FP.
        if ((TI->getOpcode() == Instruction::Sub &&
             FI->getOpcode() == Instruction::Add) ||
            (TI->getOpcode() == Instruction::FSub &&
             FI->getOpcode() == Instruction::FAdd)) {
          AddOp = FI; SubOp = TI;
        } else if ((FI->getOpcode() == Instruction::Sub &&
                    TI->getOpcode() == Instruction::Add) ||
                   (FI->getOpcode() == Instruction::FSub &&
                    TI->getOpcode() == Instruction::FAdd)) {
          AddOp = TI; SubOp = FI;
        }

        if (AddOp) {
          Value *OtherAddOp = nullptr;
          if (SubOp->getOperand(0) == AddOp->getOperand(0)) {
            OtherAddOp = AddOp->getOperand(1);
          } else if (SubOp->getOperand(0) == AddOp->getOperand(1)) {
            OtherAddOp = AddOp->getOperand(0);
          }

          if (OtherAddOp) {
            // So at this point we know we have (Y -> OtherAddOp):
            //        select C, (add X, Y), (sub X, Z)
            Value *NegVal;  // Compute -Z
            if (SI.getType()->isFPOrFPVectorTy()) {
              NegVal = Builder->CreateFNeg(SubOp->getOperand(1));
              if (Instruction *NegInst = dyn_cast<Instruction>(NegVal)) {
                FastMathFlags Flags = AddOp->getFastMathFlags();
                Flags &= SubOp->getFastMathFlags();
                NegInst->setFastMathFlags(Flags);
              }
            } else {
              NegVal = Builder->CreateNeg(SubOp->getOperand(1));
            }

            Value *NewTrueOp = OtherAddOp;
            Value *NewFalseOp = NegVal;
            if (AddOp != TI)
              std::swap(NewTrueOp, NewFalseOp);
            Value *NewSel =
              Builder->CreateSelect(CondVal, NewTrueOp,
                                    NewFalseOp, SI.getName() + ".p");

            if (SI.getType()->isFPOrFPVectorTy()) {
              Instruction *RI =
                BinaryOperator::CreateFAdd(SubOp->getOperand(0), NewSel);

              FastMathFlags Flags = AddOp->getFastMathFlags();
              Flags &= SubOp->getFastMathFlags();
              RI->setFastMathFlags(Flags);
              return RI;
            } else
              return BinaryOperator::CreateAdd(SubOp->getOperand(0), NewSel);
          }
        }
      }

  // See if we can fold the select into one of our operands.
  if (SI.getType()->isIntegerTy()) {
    if (Instruction *FoldI = FoldSelectIntoOp(SI, TrueVal, FalseVal))
      return FoldI;

    // MAX(MAX(a, b), a) -> MAX(a, b)
    // MIN(MIN(a, b), a) -> MIN(a, b)
    // MAX(MIN(a, b), a) -> a
    // MIN(MAX(a, b), a) -> a
    Value *LHS, *RHS, *LHS2, *RHS2;
    if (SelectPatternFlavor SPF = MatchSelectPattern(&SI, LHS, RHS)) {
      if (SelectPatternFlavor SPF2 = MatchSelectPattern(LHS, LHS2, RHS2))
        if (Instruction *R = FoldSPFofSPF(cast<Instruction>(LHS),SPF2,LHS2,RHS2,
                                          SI, SPF, RHS))
          return R;
      if (SelectPatternFlavor SPF2 = MatchSelectPattern(RHS, LHS2, RHS2))
        if (Instruction *R = FoldSPFofSPF(cast<Instruction>(RHS),SPF2,LHS2,RHS2,
                                          SI, SPF, LHS))
          return R;
    }

    // TODO.
    // ABS(-X) -> ABS(X)
  }

  // See if we can fold the select into a phi node if the condition is a select.
  if (isa<PHINode>(SI.getCondition()))
    // The true/false values have to be live in the PHI predecessor's blocks.
    if (CanSelectOperandBeMappingIntoPredBlock(TrueVal, SI) &&
        CanSelectOperandBeMappingIntoPredBlock(FalseVal, SI))
      if (Instruction *NV = FoldOpIntoPhi(SI))
        return NV;

  if (SelectInst *TrueSI = dyn_cast<SelectInst>(TrueVal)) {
    if (TrueSI->getCondition() == CondVal) {
      if (SI.getTrueValue() == TrueSI->getTrueValue())
        return nullptr;
      SI.setOperand(1, TrueSI->getTrueValue());
      return &SI;
    }
  }
  if (SelectInst *FalseSI = dyn_cast<SelectInst>(FalseVal)) {
    if (FalseSI->getCondition() == CondVal) {
      if (SI.getFalseValue() == FalseSI->getFalseValue())
        return nullptr;
      SI.setOperand(2, FalseSI->getFalseValue());
      return &SI;
    }
  }

  if (BinaryOperator::isNot(CondVal)) {
    SI.setOperand(0, BinaryOperator::getNotArgument(CondVal));
    SI.setOperand(1, FalseVal);
    SI.setOperand(2, TrueVal);
    return &SI;
  }

  if (VectorType* VecTy = dyn_cast<VectorType>(SI.getType())) {
    unsigned VWidth = VecTy->getNumElements();
    APInt UndefElts(VWidth, 0);
    APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
    if (Value *V = SimplifyDemandedVectorElts(&SI, AllOnesEltMask, UndefElts)) {
      if (V != &SI)
        return ReplaceInstUsesWith(SI, V);
      return &SI;
    }

    if (isa<ConstantAggregateZero>(CondVal)) {
      return ReplaceInstUsesWith(SI, FalseVal);
    }
  }

  return nullptr;
}

Instruction * InstCombiner::visitSelectInstWithICmp	(	SelectInst &	SI,
		ICmpInst *	ICI
	)

visitSelectInstWithICmp - Visit a SelectInst that has an ICmpInst as its first operand.

Definition at line 469 of file InstCombineSelect.cpp.

                                                                  {
  bool Changed = false;
  ICmpInst::Predicate Pred = ICI->getPredicate();
  Value *CmpLHS = ICI->getOperand(0);
  Value *CmpRHS = ICI->getOperand(1);
  Value *TrueVal = SI.getTrueValue();
  Value *FalseVal = SI.getFalseValue();

  // Check cases where the comparison is with a constant that
  // can be adjusted to fit the min/max idiom. We may move or edit ICI
  // here, so make sure the select is the only user.
  if (ICI->hasOneUse())
    if (ConstantInt *CI = dyn_cast<ConstantInt>(CmpRHS)) {
      // X < MIN ? T : F  -->  F
      if ((Pred == ICmpInst::ICMP_SLT || Pred == ICmpInst::ICMP_ULT)
          && CI->isMinValue(Pred == ICmpInst::ICMP_SLT))
        return ReplaceInstUsesWith(SI, FalseVal);
      // X > MAX ? T : F  -->  F
      else if ((Pred == ICmpInst::ICMP_SGT || Pred == ICmpInst::ICMP_UGT)
               && CI->isMaxValue(Pred == ICmpInst::ICMP_SGT))
        return ReplaceInstUsesWith(SI, FalseVal);
      switch (Pred) {
      default: break;
      case ICmpInst::ICMP_ULT:
      case ICmpInst::ICMP_SLT:
      case ICmpInst::ICMP_UGT:
      case ICmpInst::ICMP_SGT: {
        // These transformations only work for selects over integers.
        IntegerType *SelectTy = dyn_cast<IntegerType>(SI.getType());
        if (!SelectTy)
          break;

        Constant *AdjustedRHS;
        if (Pred == ICmpInst::ICMP_UGT || Pred == ICmpInst::ICMP_SGT)
          AdjustedRHS = ConstantInt::get(CI->getContext(), CI->getValue() + 1);
        else // (Pred == ICmpInst::ICMP_ULT || Pred == ICmpInst::ICMP_SLT)
          AdjustedRHS = ConstantInt::get(CI->getContext(), CI->getValue() - 1);

        // X > C ? X : C+1  -->  X < C+1 ? C+1 : X
        // X < C ? X : C-1  -->  X > C-1 ? C-1 : X
        if ((CmpLHS == TrueVal && AdjustedRHS == FalseVal) ||
            (CmpLHS == FalseVal && AdjustedRHS == TrueVal))
          ; // Nothing to do here. Values match without any sign/zero extension.

        // Types do not match. Instead of calculating this with mixed types
        // promote all to the larger type. This enables scalar evolution to
        // analyze this expression.
        else if (CmpRHS->getType()->getScalarSizeInBits()
                 < SelectTy->getBitWidth()) {
          Constant *sextRHS = ConstantExpr::getSExt(AdjustedRHS, SelectTy);

          // X = sext x; x >s c ? X : C+1 --> X = sext x; X <s C+1 ? C+1 : X
          // X = sext x; x <s c ? X : C-1 --> X = sext x; X >s C-1 ? C-1 : X
          // X = sext x; x >u c ? X : C+1 --> X = sext x; X <u C+1 ? C+1 : X
          // X = sext x; x <u c ? X : C-1 --> X = sext x; X >u C-1 ? C-1 : X
          if (match(TrueVal, m_SExt(m_Specific(CmpLHS))) &&
                sextRHS == FalseVal) {
            CmpLHS = TrueVal;
            AdjustedRHS = sextRHS;
          } else if (match(FalseVal, m_SExt(m_Specific(CmpLHS))) &&
                     sextRHS == TrueVal) {
            CmpLHS = FalseVal;
            AdjustedRHS = sextRHS;
          } else if (ICI->isUnsigned()) {
            Constant *zextRHS = ConstantExpr::getZExt(AdjustedRHS, SelectTy);
            // X = zext x; x >u c ? X : C+1 --> X = zext x; X <u C+1 ? C+1 : X
            // X = zext x; x <u c ? X : C-1 --> X = zext x; X >u C-1 ? C-1 : X
            // zext + signed compare cannot be changed:
            //    0xff <s 0x00, but 0x00ff >s 0x0000
            if (match(TrueVal, m_ZExt(m_Specific(CmpLHS))) &&
                zextRHS == FalseVal) {
              CmpLHS = TrueVal;
              AdjustedRHS = zextRHS;
            } else if (match(FalseVal, m_ZExt(m_Specific(CmpLHS))) &&
                       zextRHS == TrueVal) {
              CmpLHS = FalseVal;
              AdjustedRHS = zextRHS;
            } else
              break;
          } else
            break;
        } else
          break;

        Pred = ICmpInst::getSwappedPredicate(Pred);
        CmpRHS = AdjustedRHS;
        std::swap(FalseVal, TrueVal);
        ICI->setPredicate(Pred);
        ICI->setOperand(0, CmpLHS);
        ICI->setOperand(1, CmpRHS);
        SI.setOperand(1, TrueVal);
        SI.setOperand(2, FalseVal);

        // Move ICI instruction right before the select instruction. Otherwise
        // the sext/zext value may be defined after the ICI instruction uses it.
        ICI->moveBefore(&SI);

        Changed = true;
        break;
      }
      }
    }

  // Transform (X >s -1) ? C1 : C2 --> ((X >>s 31) & (C2 - C1)) + C1
  // and       (X <s  0) ? C2 : C1 --> ((X >>s 31) & (C2 - C1)) + C1
  // FIXME: Type and constness constraints could be lifted, but we have to
  //        watch code size carefully. We should consider xor instead of
  //        sub/add when we decide to do that.
  if (IntegerType *Ty = dyn_cast<IntegerType>(CmpLHS->getType())) {
    if (TrueVal->getType() == Ty) {
      if (ConstantInt *Cmp = dyn_cast<ConstantInt>(CmpRHS)) {
        ConstantInt *C1 = nullptr, *C2 = nullptr;
        if (Pred == ICmpInst::ICMP_SGT && Cmp->isAllOnesValue()) {
          C1 = dyn_cast<ConstantInt>(TrueVal);
          C2 = dyn_cast<ConstantInt>(FalseVal);
        } else if (Pred == ICmpInst::ICMP_SLT && Cmp->isNullValue()) {
          C1 = dyn_cast<ConstantInt>(FalseVal);
          C2 = dyn_cast<ConstantInt>(TrueVal);
        }
        if (C1 && C2) {
          // This shift results in either -1 or 0.
          Value *AShr = Builder->CreateAShr(CmpLHS, Ty->getBitWidth()-1);

          // Check if we can express the operation with a single or.
          if (C2->isAllOnesValue())
            return ReplaceInstUsesWith(SI, Builder->CreateOr(AShr, C1));

          Value *And = Builder->CreateAnd(AShr, C2->getValue()-C1->getValue());
          return ReplaceInstUsesWith(SI, Builder->CreateAdd(And, C1));
        }
      }
    }
  }

  // If we have an equality comparison then we know the value in one of the
  // arms of the select. See if substituting this value into the arm and
  // simplifying the result yields the same value as the other arm.
  if (Pred == ICmpInst::ICMP_EQ) {
    if (SimplifyWithOpReplaced(FalseVal, CmpLHS, CmpRHS, DL, TLI) == TrueVal ||
        SimplifyWithOpReplaced(FalseVal, CmpRHS, CmpLHS, DL, TLI) == TrueVal)
      return ReplaceInstUsesWith(SI, FalseVal);
    if (SimplifyWithOpReplaced(TrueVal, CmpLHS, CmpRHS, DL, TLI) == FalseVal ||
        SimplifyWithOpReplaced(TrueVal, CmpRHS, CmpLHS, DL, TLI) == FalseVal)
      return ReplaceInstUsesWith(SI, FalseVal);
  } else if (Pred == ICmpInst::ICMP_NE) {
    if (SimplifyWithOpReplaced(TrueVal, CmpLHS, CmpRHS, DL, TLI) == FalseVal ||
        SimplifyWithOpReplaced(TrueVal, CmpRHS, CmpLHS, DL, TLI) == FalseVal)
      return ReplaceInstUsesWith(SI, TrueVal);
    if (SimplifyWithOpReplaced(FalseVal, CmpLHS, CmpRHS, DL, TLI) == TrueVal ||
        SimplifyWithOpReplaced(FalseVal, CmpRHS, CmpLHS, DL, TLI) == TrueVal)
      return ReplaceInstUsesWith(SI, TrueVal);
  }

  // NOTE: if we wanted to, this is where to detect integer MIN/MAX

  if (CmpRHS != CmpLHS && isa<Constant>(CmpRHS)) {
    if (CmpLHS == TrueVal && Pred == ICmpInst::ICMP_EQ) {
      // Transform (X == C) ? X : Y -> (X == C) ? C : Y
      SI.setOperand(1, CmpRHS);
      Changed = true;
    } else if (CmpLHS == FalseVal && Pred == ICmpInst::ICMP_NE) {
      // Transform (X != C) ? Y : X -> (X != C) ? Y : C
      SI.setOperand(2, CmpRHS);
      Changed = true;
    }
  }

  if (Value *V = foldSelectICmpAndOr(SI, TrueVal, FalseVal, Builder))
    return ReplaceInstUsesWith(SI, V);

  return Changed ? &SI : nullptr;
}

Instruction * InstCombiner::visitSExt

(

SExtInst &

CI

)

Definition at line 1041 of file InstCombineCasts.cpp.

                                                 {
  // If this sign extend is only used by a truncate, let the truncate be
  // eliminated before we try to optimize this sext.
  if (CI.hasOneUse() && isa<TruncInst>(CI.user_back()))
    return nullptr;

  if (Instruction *I = commonCastTransforms(CI))
    return I;

  // See if we can simplify any instructions used by the input whose sole
  // purpose is to compute bits we don't care about.
  if (SimplifyDemandedInstructionBits(CI))
    return &CI;

  Value *Src = CI.getOperand(0);
  Type *SrcTy = Src->getType(), *DestTy = CI.getType();

  // Attempt to extend the entire input expression tree to the destination
  // type.   Only do this if the dest type is a simple type, don't convert the
  // expression tree to something weird like i93 unless the source is also
  // strange.
  if ((DestTy->isVectorTy() || ShouldChangeType(SrcTy, DestTy)) &&
      CanEvaluateSExtd(Src, DestTy)) {
    // Okay, we can transform this!  Insert the new expression now.
    DEBUG(dbgs() << "ICE: EvaluateInDifferentType converting expression type"
          " to avoid sign extend: " << CI);
    Value *Res = EvaluateInDifferentType(Src, DestTy, true);
    assert(Res->getType() == DestTy);

    uint32_t SrcBitSize = SrcTy->getScalarSizeInBits();
    uint32_t DestBitSize = DestTy->getScalarSizeInBits();

    // If the high bits are already filled with sign bit, just replace this
    // cast with the result.
    if (ComputeNumSignBits(Res) > DestBitSize - SrcBitSize)
      return ReplaceInstUsesWith(CI, Res);

    // We need to emit a shl + ashr to do the sign extend.
    Value *ShAmt = ConstantInt::get(DestTy, DestBitSize-SrcBitSize);
    return BinaryOperator::CreateAShr(Builder->CreateShl(Res, ShAmt, "sext"),
                                      ShAmt);
  }

  // If this input is a trunc from our destination, then turn sext(trunc(x))
  // into shifts.
  if (TruncInst *TI = dyn_cast<TruncInst>(Src))
    if (TI->hasOneUse() && TI->getOperand(0)->getType() == DestTy) {
      uint32_t SrcBitSize = SrcTy->getScalarSizeInBits();
      uint32_t DestBitSize = DestTy->getScalarSizeInBits();

      // We need to emit a shl + ashr to do the sign extend.
      Value *ShAmt = ConstantInt::get(DestTy, DestBitSize-SrcBitSize);
      Value *Res = Builder->CreateShl(TI->getOperand(0), ShAmt, "sext");
      return BinaryOperator::CreateAShr(Res, ShAmt);
    }

  if (ICmpInst *ICI = dyn_cast<ICmpInst>(Src))
    return transformSExtICmp(ICI, CI);

  // If the input is a shl/ashr pair of a same constant, then this is a sign
  // extension from a smaller value.  If we could trust arbitrary bitwidth
  // integers, we could turn this into a truncate to the smaller bit and then
  // use a sext for the whole extension.  Since we don't, look deeper and check
  // for a truncate.  If the source and dest are the same type, eliminate the
  // trunc and extend and just do shifts.  For example, turn:
  //   %a = trunc i32 %i to i8
  //   %b = shl i8 %a, 6
  //   %c = ashr i8 %b, 6
  //   %d = sext i8 %c to i32
  // into:
  //   %a = shl i32 %i, 30
  //   %d = ashr i32 %a, 30
  Value *A = nullptr;
  // TODO: Eventually this could be subsumed by EvaluateInDifferentType.
  ConstantInt *BA = nullptr, *CA = nullptr;
  if (match(Src, m_AShr(m_Shl(m_Trunc(m_Value(A)), m_ConstantInt(BA)),
                        m_ConstantInt(CA))) &&
      BA == CA && A->getType() == CI.getType()) {
    unsigned MidSize = Src->getType()->getScalarSizeInBits();
    unsigned SrcDstSize = CI.getType()->getScalarSizeInBits();
    unsigned ShAmt = CA->getZExtValue()+SrcDstSize-MidSize;
    Constant *ShAmtV = ConstantInt::get(CI.getType(), ShAmt);
    A = Builder->CreateShl(A, ShAmtV, CI.getName());
    return BinaryOperator::CreateAShr(A, ShAmtV);
  }

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitSExtInst ( SExtInst &  I )

inlineinherited

Definition at line 184 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitShl

(

BinaryOperator &

I

)

Definition at line 688 of file InstCombineShifts.cpp.

                                                     {
  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyShlInst(I.getOperand(0), I.getOperand(1),
                                 I.hasNoSignedWrap(), I.hasNoUnsignedWrap(),
                                 DL))
    return ReplaceInstUsesWith(I, V);

  if (Instruction *V = commonShiftTransforms(I))
    return V;

  if (ConstantInt *Op1C = dyn_cast<ConstantInt>(I.getOperand(1))) {
    unsigned ShAmt = Op1C->getZExtValue();

    // If the shifted-out value is known-zero, then this is a NUW shift.
    if (!I.hasNoUnsignedWrap() &&
        MaskedValueIsZero(I.getOperand(0),
                          APInt::getHighBitsSet(Op1C->getBitWidth(), ShAmt))) {
          I.setHasNoUnsignedWrap();
          return &I;
        }

    // If the shifted out value is all signbits, this is a NSW shift.
    if (!I.hasNoSignedWrap() &&
        ComputeNumSignBits(I.getOperand(0)) > ShAmt) {
      I.setHasNoSignedWrap();
      return &I;
    }
  }

  // (C1 << A) << C2 -> (C1 << C2) << A
  Constant *C1, *C2;
  Value *A;
  if (match(I.getOperand(0), m_OneUse(m_Shl(m_Constant(C1), m_Value(A)))) &&
      match(I.getOperand(1), m_Constant(C2)))
    return BinaryOperator::CreateShl(ConstantExpr::getShl(C1, C2), A);

  return nullptr;
}

Instruction * InstCombiner::visitShuffleVectorInst

(

ShuffleVectorInst &

SVI

)

Definition at line 856 of file InstCombineVectorOps.cpp.

                                                                        {
  Value *LHS = SVI.getOperand(0);
  Value *RHS = SVI.getOperand(1);
  SmallVector<int, 16> Mask = SVI.getShuffleMask();

  bool MadeChange = false;

  // Undefined shuffle mask -> undefined value.
  if (isa<UndefValue>(SVI.getOperand(2)))
    return ReplaceInstUsesWith(SVI, UndefValue::get(SVI.getType()));

  unsigned VWidth = cast<VectorType>(SVI.getType())->getNumElements();

  APInt UndefElts(VWidth, 0);
  APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
  if (Value *V = SimplifyDemandedVectorElts(&SVI, AllOnesEltMask, UndefElts)) {
    if (V != &SVI)
      return ReplaceInstUsesWith(SVI, V);
    LHS = SVI.getOperand(0);
    RHS = SVI.getOperand(1);
    MadeChange = true;
  }

  unsigned LHSWidth = cast<VectorType>(LHS->getType())->getNumElements();

  // Canonicalize shuffle(x    ,x,mask) -> shuffle(x, undef,mask')
  // Canonicalize shuffle(undef,x,mask) -> shuffle(x, undef,mask').
  if (LHS == RHS || isa<UndefValue>(LHS)) {
    if (isa<UndefValue>(LHS) && LHS == RHS) {
      // shuffle(undef,undef,mask) -> undef.
      Value *Result = (VWidth == LHSWidth)
                      ? LHS : UndefValue::get(SVI.getType());
      return ReplaceInstUsesWith(SVI, Result);
    }

    // Remap any references to RHS to use LHS.
    SmallVector<Constant*, 16> Elts;
    for (unsigned i = 0, e = LHSWidth; i != VWidth; ++i) {
      if (Mask[i] < 0) {
        Elts.push_back(UndefValue::get(Type::getInt32Ty(SVI.getContext())));
        continue;
      }

      if ((Mask[i] >= (int)e && isa<UndefValue>(RHS)) ||
          (Mask[i] <  (int)e && isa<UndefValue>(LHS))) {
        Mask[i] = -1;     // Turn into undef.
        Elts.push_back(UndefValue::get(Type::getInt32Ty(SVI.getContext())));
      } else {
        Mask[i] = Mask[i] % e;  // Force to LHS.
        Elts.push_back(ConstantInt::get(Type::getInt32Ty(SVI.getContext()),
                                        Mask[i]));
      }
    }
    SVI.setOperand(0, SVI.getOperand(1));
    SVI.setOperand(1, UndefValue::get(RHS->getType()));
    SVI.setOperand(2, ConstantVector::get(Elts));
    LHS = SVI.getOperand(0);
    RHS = SVI.getOperand(1);
    MadeChange = true;
  }

  if (VWidth == LHSWidth) {
    // Analyze the shuffle, are the LHS or RHS and identity shuffles?
    bool isLHSID, isRHSID;
    RecognizeIdentityMask(Mask, isLHSID, isRHSID);

    // Eliminate identity shuffles.
    if (isLHSID) return ReplaceInstUsesWith(SVI, LHS);
    if (isRHSID) return ReplaceInstUsesWith(SVI, RHS);
  }

  if (isa<UndefValue>(RHS) && CanEvaluateShuffled(LHS, Mask)) {
    Value *V = EvaluateInDifferentElementOrder(LHS, Mask);
    return ReplaceInstUsesWith(SVI, V);
  }

  // If the LHS is a shufflevector itself, see if we can combine it with this
  // one without producing an unusual shuffle.
  // Cases that might be simplified:
  // 1.
  // x1=shuffle(v1,v2,mask1)
  //  x=shuffle(x1,undef,mask)
  //        ==>
  //  x=shuffle(v1,undef,newMask)
  // newMask[i] = (mask[i] < x1.size()) ? mask1[mask[i]] : -1
  // 2.
  // x1=shuffle(v1,undef,mask1)
  //  x=shuffle(x1,x2,mask)
  // where v1.size() == mask1.size()
  //        ==>
  //  x=shuffle(v1,x2,newMask)
  // newMask[i] = (mask[i] < x1.size()) ? mask1[mask[i]] : mask[i]
  // 3.
  // x2=shuffle(v2,undef,mask2)
  //  x=shuffle(x1,x2,mask)
  // where v2.size() == mask2.size()
  //        ==>
  //  x=shuffle(x1,v2,newMask)
  // newMask[i] = (mask[i] < x1.size())
  //              ? mask[i] : mask2[mask[i]-x1.size()]+x1.size()
  // 4.
  // x1=shuffle(v1,undef,mask1)
  // x2=shuffle(v2,undef,mask2)
  //  x=shuffle(x1,x2,mask)
  // where v1.size() == v2.size()
  //        ==>
  //  x=shuffle(v1,v2,newMask)
  // newMask[i] = (mask[i] < x1.size())
  //              ? mask1[mask[i]] : mask2[mask[i]-x1.size()]+v1.size()
  //
  // Here we are really conservative:
  // we are absolutely afraid of producing a shuffle mask not in the input
  // program, because the code gen may not be smart enough to turn a merged
  // shuffle into two specific shuffles: it may produce worse code.  As such,
  // we only merge two shuffles if the result is either a splat or one of the
  // input shuffle masks.  In this case, merging the shuffles just removes
  // one instruction, which we know is safe.  This is good for things like
  // turning: (splat(splat)) -> splat, or
  // merge(V[0..n], V[n+1..2n]) -> V[0..2n]
  ShuffleVectorInst* LHSShuffle = dyn_cast<ShuffleVectorInst>(LHS);
  ShuffleVectorInst* RHSShuffle = dyn_cast<ShuffleVectorInst>(RHS);
  if (LHSShuffle)
    if (!isa<UndefValue>(LHSShuffle->getOperand(1)) && !isa<UndefValue>(RHS))
      LHSShuffle = nullptr;
  if (RHSShuffle)
    if (!isa<UndefValue>(RHSShuffle->getOperand(1)))
      RHSShuffle = nullptr;
  if (!LHSShuffle && !RHSShuffle)
    return MadeChange ? &SVI : nullptr;

  Value* LHSOp0 = nullptr;
  Value* LHSOp1 = nullptr;
  Value* RHSOp0 = nullptr;
  unsigned LHSOp0Width = 0;
  unsigned RHSOp0Width = 0;
  if (LHSShuffle) {
    LHSOp0 = LHSShuffle->getOperand(0);
    LHSOp1 = LHSShuffle->getOperand(1);
    LHSOp0Width = cast<VectorType>(LHSOp0->getType())->getNumElements();
  }
  if (RHSShuffle) {
    RHSOp0 = RHSShuffle->getOperand(0);
    RHSOp0Width = cast<VectorType>(RHSOp0->getType())->getNumElements();
  }
  Value* newLHS = LHS;
  Value* newRHS = RHS;
  if (LHSShuffle) {
    // case 1
    if (isa<UndefValue>(RHS)) {
      newLHS = LHSOp0;
      newRHS = LHSOp1;
    }
    // case 2 or 4
    else if (LHSOp0Width == LHSWidth) {
      newLHS = LHSOp0;
    }
  }
  // case 3 or 4
  if (RHSShuffle && RHSOp0Width == LHSWidth) {
    newRHS = RHSOp0;
  }
  // case 4
  if (LHSOp0 == RHSOp0) {
    newLHS = LHSOp0;
    newRHS = nullptr;
  }

  if (newLHS == LHS && newRHS == RHS)
    return MadeChange ? &SVI : nullptr;

  SmallVector<int, 16> LHSMask;
  SmallVector<int, 16> RHSMask;
  if (newLHS != LHS)
    LHSMask = LHSShuffle->getShuffleMask();
  if (RHSShuffle && newRHS != RHS)
    RHSMask = RHSShuffle->getShuffleMask();

  unsigned newLHSWidth = (newLHS != LHS) ? LHSOp0Width : LHSWidth;
  SmallVector<int, 16> newMask;
  bool isSplat = true;
  int SplatElt = -1;
  // Create a new mask for the new ShuffleVectorInst so that the new
  // ShuffleVectorInst is equivalent to the original one.
  for (unsigned i = 0; i < VWidth; ++i) {
    int eltMask;
    if (Mask[i] < 0) {
      // This element is an undef value.
      eltMask = -1;
    } else if (Mask[i] < (int)LHSWidth) {
      // This element is from left hand side vector operand.
      //
      // If LHS is going to be replaced (case 1, 2, or 4), calculate the
      // new mask value for the element.
      if (newLHS != LHS) {
        eltMask = LHSMask[Mask[i]];
        // If the value selected is an undef value, explicitly specify it
        // with a -1 mask value.
        if (eltMask >= (int)LHSOp0Width && isa<UndefValue>(LHSOp1))
          eltMask = -1;
      } else
        eltMask = Mask[i];
    } else {
      // This element is from right hand side vector operand
      //
      // If the value selected is an undef value, explicitly specify it
      // with a -1 mask value. (case 1)
      if (isa<UndefValue>(RHS))
        eltMask = -1;
      // If RHS is going to be replaced (case 3 or 4), calculate the
      // new mask value for the element.
      else if (newRHS != RHS) {
        eltMask = RHSMask[Mask[i]-LHSWidth];
        // If the value selected is an undef value, explicitly specify it
        // with a -1 mask value.
        if (eltMask >= (int)RHSOp0Width) {
          assert(isa<UndefValue>(RHSShuffle->getOperand(1))
                 && "should have been check above");
          eltMask = -1;
        }
      } else
        eltMask = Mask[i]-LHSWidth;

      // If LHS's width is changed, shift the mask value accordingly.
      // If newRHS == NULL, i.e. LHSOp0 == RHSOp0, we want to remap any
      // references from RHSOp0 to LHSOp0, so we don't need to shift the mask.
      // If newRHS == newLHS, we want to remap any references from newRHS to
      // newLHS so that we can properly identify splats that may occur due to
      // obfuscation across the two vectors.
      if (eltMask >= 0 && newRHS != nullptr && newLHS != newRHS)
        eltMask += newLHSWidth;
    }

    // Check if this could still be a splat.
    if (eltMask >= 0) {
      if (SplatElt >= 0 && SplatElt != eltMask)
        isSplat = false;
      SplatElt = eltMask;
    }

    newMask.push_back(eltMask);
  }

  // If the result mask is equal to one of the original shuffle masks,
  // or is a splat, do the replacement.
  if (isSplat || newMask == LHSMask || newMask == RHSMask || newMask == Mask) {
    SmallVector<Constant*, 16> Elts;
    Type *Int32Ty = Type::getInt32Ty(SVI.getContext());
    for (unsigned i = 0, e = newMask.size(); i != e; ++i) {
      if (newMask[i] < 0) {
        Elts.push_back(UndefValue::get(Int32Ty));
      } else {
        Elts.push_back(ConstantInt::get(Int32Ty, newMask[i]));
      }
    }
    if (!newRHS)
      newRHS = UndefValue::get(newLHS->getType());
    return new ShuffleVectorInst(newLHS, newRHS, ConstantVector::get(Elts));
  }

  // If the result mask is an identity, replace uses of this instruction with
  // corresponding argument.
  bool isLHSID, isRHSID;
  RecognizeIdentityMask(newMask, isLHSID, isRHSID);
  if (isLHSID && VWidth == LHSOp0Width) return ReplaceInstUsesWith(SVI, newLHS);
  if (isRHSID && VWidth == RHSOp0Width) return ReplaceInstUsesWith(SVI, newRHS);

  return MadeChange ? &SVI : nullptr;
}

Instruction * InstCombiner::visitSIToFP

(

CastInst &

CI

)

Definition at line 1402 of file InstCombineCasts.cpp.

                                                   {
  return commonCastTransforms(CI);
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitSIToFPInst ( SIToFPInst &  I )

inlineinherited

Definition at line 190 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitSRem

(

BinaryOperator &

I

)

Definition at line 1261 of file InstCombineMulDivRem.cpp.

                                                      {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifySRemInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // Handle the integer rem common cases
  if (Instruction *Common = commonIRemTransforms(I))
    return Common;

  if (Value *RHSNeg = dyn_castNegVal(Op1))
    if (!isa<Constant>(RHSNeg) ||
        (isa<ConstantInt>(RHSNeg) &&
         cast<ConstantInt>(RHSNeg)->getValue().isStrictlyPositive())) {
      // X % -Y -> X % Y
      Worklist.AddValue(I.getOperand(1));
      I.setOperand(1, RHSNeg);
      return &I;
    }

  // If the sign bits of both operands are zero (i.e. we can prove they are
  // unsigned inputs), turn this into a urem.
  if (I.getType()->isIntegerTy()) {
    APInt Mask(APInt::getSignBit(I.getType()->getPrimitiveSizeInBits()));
    if (MaskedValueIsZero(Op1, Mask) && MaskedValueIsZero(Op0, Mask)) {
      // X srem Y -> X urem Y, iff X and Y don't have sign bit set
      return BinaryOperator::CreateURem(Op0, Op1, I.getName());
    }
  }

  // If it's a constant vector, flip any negative values positive.
  if (isa<ConstantVector>(Op1) || isa<ConstantDataVector>(Op1)) {
    Constant *C = cast<Constant>(Op1);
    unsigned VWidth = C->getType()->getVectorNumElements();

    bool hasNegative = false;
    bool hasMissing = false;
    for (unsigned i = 0; i != VWidth; ++i) {
      Constant *Elt = C->getAggregateElement(i);
      if (!Elt) {
        hasMissing = true;
        break;
      }

      if (ConstantInt *RHS = dyn_cast<ConstantInt>(Elt))
        if (RHS->isNegative())
          hasNegative = true;
    }

    if (hasNegative && !hasMissing) {
      SmallVector<Constant *, 16> Elts(VWidth);
      for (unsigned i = 0; i != VWidth; ++i) {
        Elts[i] = C->getAggregateElement(i);  // Handle undef, etc.
        if (ConstantInt *RHS = dyn_cast<ConstantInt>(Elts[i])) {
          if (RHS->isNegative())
            Elts[i] = cast<ConstantInt>(ConstantExpr::getNeg(RHS));
        }
      }

      Constant *NewRHSV = ConstantVector::get(Elts);
      if (NewRHSV != C) {  // Don't loop on -MININT
        Worklist.AddValue(I.getOperand(1));
        I.setOperand(1, NewRHSV);
        return &I;
      }
    }
  }

  return nullptr;
}

Instruction * InstCombiner::visitStoreInst

(

StoreInst &

SI

)

Definition at line 596 of file InstCombineLoadStoreAlloca.cpp.

                                                       {
  Value *Val = SI.getOperand(0);
  Value *Ptr = SI.getOperand(1);

  // Attempt to improve the alignment.
  if (DL) {
    unsigned KnownAlign =
      getOrEnforceKnownAlignment(Ptr, DL->getPrefTypeAlignment(Val->getType()),
                                 DL);
    unsigned StoreAlign = SI.getAlignment();
    unsigned EffectiveStoreAlign = StoreAlign != 0 ? StoreAlign :
      DL->getABITypeAlignment(Val->getType());

    if (KnownAlign > EffectiveStoreAlign)
      SI.setAlignment(KnownAlign);
    else if (StoreAlign == 0)
      SI.setAlignment(EffectiveStoreAlign);
  }

  // Don't hack volatile/atomic stores.
  // FIXME: Some bits are legal for atomic stores; needs refactoring.
  if (!SI.isSimple()) return nullptr;

  // If the RHS is an alloca with a single use, zapify the store, making the
  // alloca dead.
  if (Ptr->hasOneUse()) {
    if (isa<AllocaInst>(Ptr))
      return EraseInstFromFunction(SI);
    if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr)) {
      if (isa<AllocaInst>(GEP->getOperand(0))) {
        if (GEP->getOperand(0)->hasOneUse())
          return EraseInstFromFunction(SI);
      }
    }
  }

  // Do really simple DSE, to catch cases where there are several consecutive
  // stores to the same location, separated by a few arithmetic operations. This
  // situation often occurs with bitfield accesses.
  BasicBlock::iterator BBI = &SI;
  for (unsigned ScanInsts = 6; BBI != SI.getParent()->begin() && ScanInsts;
       --ScanInsts) {
    --BBI;
    // Don't count debug info directives, lest they affect codegen,
    // and we skip pointer-to-pointer bitcasts, which are NOPs.
    if (isa<DbgInfoIntrinsic>(BBI) ||
        (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
      ScanInsts++;
      continue;
    }

    if (StoreInst *PrevSI = dyn_cast<StoreInst>(BBI)) {
      // Prev store isn't volatile, and stores to the same location?
      if (PrevSI->isSimple() && equivalentAddressValues(PrevSI->getOperand(1),
                                                        SI.getOperand(1))) {
        ++NumDeadStore;
        ++BBI;
        EraseInstFromFunction(*PrevSI);
        continue;
      }
      break;
    }

    // If this is a load, we have to stop.  However, if the loaded value is from
    // the pointer we're loading and is producing the pointer we're storing,
    // then *this* store is dead (X = load P; store X -> P).
    if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
      if (LI == Val && equivalentAddressValues(LI->getOperand(0), Ptr) &&
          LI->isSimple())
        return EraseInstFromFunction(SI);

      // Otherwise, this is a load from some other location.  Stores before it
      // may not be dead.
      break;
    }

    // Don't skip over loads or things that can modify memory.
    if (BBI->mayWriteToMemory() || BBI->mayReadFromMemory())
      break;
  }

  // store X, null    -> turns into 'unreachable' in SimplifyCFG
  if (isa<ConstantPointerNull>(Ptr) && SI.getPointerAddressSpace() == 0) {
    if (!isa<UndefValue>(Val)) {
      SI.setOperand(0, UndefValue::get(Val->getType()));
      if (Instruction *U = dyn_cast<Instruction>(Val))
        Worklist.Add(U);  // Dropped a use.
    }
    return nullptr;  // Do not modify these!
  }

  // store undef, Ptr -> noop
  if (isa<UndefValue>(Val))
    return EraseInstFromFunction(SI);

  // If the pointer destination is a cast, see if we can fold the cast into the
  // source instead.
  if (isa<CastInst>(Ptr))
    if (Instruction *Res = InstCombineStoreToCast(*this, SI))
      return Res;
  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr))
    if (CE->isCast())
      if (Instruction *Res = InstCombineStoreToCast(*this, SI))
        return Res;


  // If this store is the last instruction in the basic block (possibly
  // excepting debug info instructions), and if the block ends with an
  // unconditional branch, try to move it to the successor block.
  BBI = &SI;
  do {
    ++BBI;
  } while (isa<DbgInfoIntrinsic>(BBI) ||
           (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy()));
  if (BranchInst *BI = dyn_cast<BranchInst>(BBI))
    if (BI->isUnconditional())
      if (SimplifyStoreAtEndOfBlock(SI))
        return nullptr;  // xform done!

  return nullptr;
}

Instruction * InstCombiner::visitSub

(

BinaryOperator &

I

)

Definition at line 1451 of file InstCombineAddSub.cpp.

                                                     {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifySubInst(Op0, Op1, I.hasNoSignedWrap(),
                                 I.hasNoUnsignedWrap(), DL))
    return ReplaceInstUsesWith(I, V);

  // (A*B)-(A*C) -> A*(B-C) etc
  if (Value *V = SimplifyUsingDistributiveLaws(I))
    return ReplaceInstUsesWith(I, V);

  // If this is a 'B = x-(-A)', change to B = x+A.  This preserves NSW/NUW.
  if (Value *V = dyn_castNegVal(Op1)) {
    BinaryOperator *Res = BinaryOperator::CreateAdd(Op0, V);
    Res->setHasNoSignedWrap(I.hasNoSignedWrap());
    Res->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
    return Res;
  }

  if (I.getType()->isIntegerTy(1))
    return BinaryOperator::CreateXor(Op0, Op1);

  // Replace (-1 - A) with (~A).
  if (match(Op0, m_AllOnes()))
    return BinaryOperator::CreateNot(Op1);

  if (Constant *C = dyn_cast<Constant>(Op0)) {
    // C - ~X == X + (1+C)
    Value *X = nullptr;
    if (match(Op1, m_Not(m_Value(X))))
      return BinaryOperator::CreateAdd(X, AddOne(C));

    // Try to fold constant sub into select arguments.
    if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;

    // C-(X+C2) --> (C-C2)-X
    Constant *C2;
    if (match(Op1, m_Add(m_Value(X), m_Constant(C2))))
      return BinaryOperator::CreateSub(ConstantExpr::getSub(C, C2), X);

    if (SimplifyDemandedInstructionBits(I))
      return &I;

    // Fold (sub 0, (zext bool to B)) --> (sext bool to B)
    if (C->isNullValue() && match(Op1, m_ZExt(m_Value(X))))
      if (X->getType()->getScalarType()->isIntegerTy(1))
        return CastInst::CreateSExtOrBitCast(X, Op1->getType());

    // Fold (sub 0, (sext bool to B)) --> (zext bool to B)
    if (C->isNullValue() && match(Op1, m_SExt(m_Value(X))))
      if (X->getType()->getScalarType()->isIntegerTy(1))
        return CastInst::CreateZExtOrBitCast(X, Op1->getType());
  }

  if (ConstantInt *C = dyn_cast<ConstantInt>(Op0)) {
    // -(X >>u 31) -> (X >>s 31)
    // -(X >>s 31) -> (X >>u 31)
    if (C->isZero()) {
      Value *X; ConstantInt *CI;
      if (match(Op1, m_LShr(m_Value(X), m_ConstantInt(CI))) &&
          // Verify we are shifting out everything but the sign bit.
          CI->getValue() == I.getType()->getPrimitiveSizeInBits()-1)
        return BinaryOperator::CreateAShr(X, CI);

      if (match(Op1, m_AShr(m_Value(X), m_ConstantInt(CI))) &&
          // Verify we are shifting out everything but the sign bit.
          CI->getValue() == I.getType()->getPrimitiveSizeInBits()-1)
        return BinaryOperator::CreateLShr(X, CI);
    }
  }


  { Value *Y;
    // X-(X+Y) == -Y    X-(Y+X) == -Y
    if (match(Op1, m_Add(m_Specific(Op0), m_Value(Y))) ||
        match(Op1, m_Add(m_Value(Y), m_Specific(Op0))))
      return BinaryOperator::CreateNeg(Y);

    // (X-Y)-X == -Y
    if (match(Op0, m_Sub(m_Specific(Op1), m_Value(Y))))
      return BinaryOperator::CreateNeg(Y);
  }

  if (Op1->hasOneUse()) {
    Value *X = nullptr, *Y = nullptr, *Z = nullptr;
    Constant *C = nullptr;
    Constant *CI = nullptr;

    // (X - (Y - Z))  -->  (X + (Z - Y)).
    if (match(Op1, m_Sub(m_Value(Y), m_Value(Z))))
      return BinaryOperator::CreateAdd(Op0,
                                      Builder->CreateSub(Z, Y, Op1->getName()));

    // (X - (X & Y))   -->   (X & ~Y)
    //
    if (match(Op1, m_And(m_Value(Y), m_Specific(Op0))) ||
        match(Op1, m_And(m_Specific(Op0), m_Value(Y))))
      return BinaryOperator::CreateAnd(Op0,
                                  Builder->CreateNot(Y, Y->getName() + ".not"));

    // 0 - (X sdiv C)  -> (X sdiv -C)  provided the negation doesn't overflow.
    if (match(Op1, m_SDiv(m_Value(X), m_Constant(C))) && match(Op0, m_Zero()) &&
        !C->isMinSignedValue())
      return BinaryOperator::CreateSDiv(X, ConstantExpr::getNeg(C));

    // 0 - (X << Y)  -> (-X << Y)   when X is freely negatable.
    if (match(Op1, m_Shl(m_Value(X), m_Value(Y))) && match(Op0, m_Zero()))
      if (Value *XNeg = dyn_castNegVal(X))
        return BinaryOperator::CreateShl(XNeg, Y);

    // X - A*-B -> X + A*B
    // X - -A*B -> X + A*B
    Value *A, *B;
    if (match(Op1, m_Mul(m_Value(A), m_Neg(m_Value(B)))) ||
        match(Op1, m_Mul(m_Neg(m_Value(A)), m_Value(B))))
      return BinaryOperator::CreateAdd(Op0, Builder->CreateMul(A, B));

    // X - A*CI -> X + A*-CI
    // X - CI*A -> X + A*-CI
    if (match(Op1, m_Mul(m_Value(A), m_Constant(CI))) ||
        match(Op1, m_Mul(m_Constant(CI), m_Value(A)))) {
      Value *NewMul = Builder->CreateMul(A, ConstantExpr::getNeg(CI));
      return BinaryOperator::CreateAdd(Op0, NewMul);
    }
  }

  // Optimize pointer differences into the same array into a size.  Consider:
  //  &A[10] - &A[0]: we should compile this to "10".
  if (DL) {
    Value *LHSOp, *RHSOp;
    if (match(Op0, m_PtrToInt(m_Value(LHSOp))) &&
        match(Op1, m_PtrToInt(m_Value(RHSOp))))
      if (Value *Res = OptimizePointerDifference(LHSOp, RHSOp, I.getType()))
        return ReplaceInstUsesWith(I, Res);

    // trunc(p)-trunc(q) -> trunc(p-q)
    if (match(Op0, m_Trunc(m_PtrToInt(m_Value(LHSOp)))) &&
        match(Op1, m_Trunc(m_PtrToInt(m_Value(RHSOp)))))
      if (Value *Res = OptimizePointerDifference(LHSOp, RHSOp, I.getType()))
        return ReplaceInstUsesWith(I, Res);
      }

  return nullptr;
}

Instruction * InstCombiner::visitSwitchInst

(

SwitchInst &

SI

)

Definition at line 1975 of file InstructionCombining.cpp.

                                                         {
  Value *Cond = SI.getCondition();
  if (Instruction *I = dyn_cast<Instruction>(Cond)) {
    if (I->getOpcode() == Instruction::Add)
      if (ConstantInt *AddRHS = dyn_cast<ConstantInt>(I->getOperand(1))) {
        // change 'switch (X+4) case 1:' into 'switch (X) case -3'
        // Skip the first item since that's the default case.
        for (SwitchInst::CaseIt i = SI.case_begin(), e = SI.case_end();
             i != e; ++i) {
          ConstantInt* CaseVal = i.getCaseValue();
          Constant* NewCaseVal = ConstantExpr::getSub(cast<Constant>(CaseVal),
                                                      AddRHS);
          assert(isa<ConstantInt>(NewCaseVal) &&
                 "Result of expression should be constant");
          i.setValue(cast<ConstantInt>(NewCaseVal));
        }
        SI.setCondition(I->getOperand(0));
        Worklist.Add(I);
        return &SI;
      }
  }
  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitTerminatorInst ( TerminatorInst &  I )

inlineinherited

Definition at line 234 of file InstVisitor.h.

{ DELEGATE(Instruction);}

Instruction * InstCombiner::visitTrunc

(

TruncInst &

CI

)

Definition at line 439 of file InstCombineCasts.cpp.

                                                   {
  if (Instruction *Result = commonCastTransforms(CI))
    return Result;

  // See if we can simplify any instructions used by the input whose sole
  // purpose is to compute bits we don't care about.
  if (SimplifyDemandedInstructionBits(CI))
    return &CI;

  Value *Src = CI.getOperand(0);
  Type *DestTy = CI.getType(), *SrcTy = Src->getType();

  // Attempt to truncate the entire input expression tree to the destination
  // type.   Only do this if the dest type is a simple type, don't convert the
  // expression tree to something weird like i93 unless the source is also
  // strange.
  if ((DestTy->isVectorTy() || ShouldChangeType(SrcTy, DestTy)) &&
      CanEvaluateTruncated(Src, DestTy)) {

    // If this cast is a truncate, evaluting in a different type always
    // eliminates the cast, so it is always a win.
    DEBUG(dbgs() << "ICE: EvaluateInDifferentType converting expression type"
          " to avoid cast: " << CI << '\n');
    Value *Res = EvaluateInDifferentType(Src, DestTy, false);
    assert(Res->getType() == DestTy);
    return ReplaceInstUsesWith(CI, Res);
  }

  // Canonicalize trunc x to i1 -> (icmp ne (and x, 1), 0), likewise for vector.
  if (DestTy->getScalarSizeInBits() == 1) {
    Constant *One = ConstantInt::get(Src->getType(), 1);
    Src = Builder->CreateAnd(Src, One);
    Value *Zero = Constant::getNullValue(Src->getType());
    return new ICmpInst(ICmpInst::ICMP_NE, Src, Zero);
  }

  // Transform trunc(lshr (zext A), Cst) to eliminate one type conversion.
  Value *A = nullptr; ConstantInt *Cst = nullptr;
  if (Src->hasOneUse() &&
      match(Src, m_LShr(m_ZExt(m_Value(A)), m_ConstantInt(Cst)))) {
    // We have three types to worry about here, the type of A, the source of
    // the truncate (MidSize), and the destination of the truncate. We know that
    // ASize < MidSize   and MidSize > ResultSize, but don't know the relation
    // between ASize and ResultSize.
    unsigned ASize = A->getType()->getPrimitiveSizeInBits();

    // If the shift amount is larger than the size of A, then the result is
    // known to be zero because all the input bits got shifted out.
    if (Cst->getZExtValue() >= ASize)
      return ReplaceInstUsesWith(CI, Constant::getNullValue(CI.getType()));

    // Since we're doing an lshr and a zero extend, and know that the shift
    // amount is smaller than ASize, it is always safe to do the shift in A's
    // type, then zero extend or truncate to the result.
    Value *Shift = Builder->CreateLShr(A, Cst->getZExtValue());
    Shift->takeName(Src);
    return CastInst::CreateIntegerCast(Shift, CI.getType(), false);
  }

  // Transform "trunc (and X, cst)" -> "and (trunc X), cst" so long as the dest
  // type isn't non-native.
  if (Src->hasOneUse() && isa<IntegerType>(Src->getType()) &&
      ShouldChangeType(Src->getType(), CI.getType()) &&
      match(Src, m_And(m_Value(A), m_ConstantInt(Cst)))) {
    Value *NewTrunc = Builder->CreateTrunc(A, CI.getType(), A->getName()+".tr");
    return BinaryOperator::CreateAnd(NewTrunc,
                                     ConstantExpr::getTrunc(Cst, CI.getType()));
  }

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitTruncInst ( TruncInst &  I )

inlineinherited

Definition at line 182 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * InstCombiner::visitUDiv

(

BinaryOperator &

I

)

Definition at line 906 of file InstCombineMulDivRem.cpp.

                                                      {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyUDivInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // Handle the integer div common cases
  if (Instruction *Common = commonIDivTransforms(I))
    return Common;

  // (x lshr C1) udiv C2 --> x udiv (C2 << C1)
  if (Constant *C2 = dyn_cast<Constant>(Op1)) {
    Value *X;
    Constant *C1;
    if (match(Op0, m_LShr(m_Value(X), m_Constant(C1))))
      return BinaryOperator::CreateUDiv(X, ConstantExpr::getShl(C2, C1));
  }

  // (zext A) udiv (zext B) --> zext (A udiv B)
  if (ZExtInst *ZOp0 = dyn_cast<ZExtInst>(Op0))
    if (Value *ZOp1 = dyn_castZExtVal(Op1, ZOp0->getSrcTy()))
      return new ZExtInst(Builder->CreateUDiv(ZOp0->getOperand(0), ZOp1, "div",
                                              I.isExact()),
                          I.getType());

  // (LHS udiv (select (select (...)))) -> (LHS >> (select (select (...))))
  SmallVector<UDivFoldAction, 6> UDivActions;
  if (visitUDivOperand(Op0, Op1, I, UDivActions))
    for (unsigned i = 0, e = UDivActions.size(); i != e; ++i) {
      FoldUDivOperandCb Action = UDivActions[i].FoldAction;
      Value *ActionOp1 = UDivActions[i].OperandToFold;
      Instruction *Inst;
      if (Action)
        Inst = Action(Op0, ActionOp1, I, *this);
      else {
        // This action joins two actions together.  The RHS of this action is
        // simply the last action we processed, we saved the LHS action index in
        // the joining action.
        size_t SelectRHSIdx = i - 1;
        Value *SelectRHS = UDivActions[SelectRHSIdx].FoldResult;
        size_t SelectLHSIdx = UDivActions[i].SelectLHSIdx;
        Value *SelectLHS = UDivActions[SelectLHSIdx].FoldResult;
        Inst = SelectInst::Create(cast<SelectInst>(ActionOp1)->getCondition(),
                                  SelectLHS, SelectRHS);
      }

      // If this is the last action to process, return it to the InstCombiner.
      // Otherwise, we insert it before the UDiv and record it so that we may
      // use it as part of a joining action (i.e., a SelectInst).
      if (e - i != 1) {
        Inst->insertBefore(&I);
        UDivActions[i].FoldResult = Inst;
      } else
        return Inst;
    }

  return nullptr;
}

Instruction * InstCombiner::visitUIToFP

(

CastInst &

CI

)

Definition at line 1398 of file InstCombineCasts.cpp.

                                                   {
  return commonCastTransforms(CI);
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitUIToFPInst ( UIToFPInst &  I )

inlineinherited

Definition at line 189 of file InstVisitor.h.

{ DELEGATE(CastInst);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitUnaryInstruction ( UnaryInstruction &  I )

inlineinherited

Definition at line 235 of file InstVisitor.h.

{ DELEGATE(Instruction);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitUnreachableInst ( UnreachableInst &  I )

inlineinherited

Definition at line 171 of file InstVisitor.h.

{ DELEGATE(TerminatorInst);}

Instruction * InstCombiner::visitURem

(

BinaryOperator &

I

)

Definition at line 1226 of file InstCombineMulDivRem.cpp.

                                                      {
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyURemInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  if (Instruction *common = commonIRemTransforms(I))
    return common;

  // (zext A) urem (zext B) --> zext (A urem B)
  if (ZExtInst *ZOp0 = dyn_cast<ZExtInst>(Op0))
    if (Value *ZOp1 = dyn_castZExtVal(Op1, ZOp0->getSrcTy()))
      return new ZExtInst(Builder->CreateURem(ZOp0->getOperand(0), ZOp1),
                          I.getType());

  // X urem Y -> X and Y-1, where Y is a power of 2,
  if (isKnownToBeAPowerOfTwo(Op1, /*OrZero*/true)) {
    Constant *N1 = Constant::getAllOnesValue(I.getType());
    Value *Add = Builder->CreateAdd(Op1, N1);
    return BinaryOperator::CreateAnd(Op0, Add);
  }

  // 1 urem X -> zext(X != 1)
  if (match(Op0, m_One())) {
    Value *Cmp = Builder->CreateICmpNE(Op1, Op0);
    Value *Ext = Builder->CreateZExt(Cmp, I.getType());
    return ReplaceInstUsesWith(I, Ext);
  }

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitVAArgInst ( VAArgInst &  I )

inlineinherited

Definition at line 196 of file InstVisitor.h.

{ DELEGATE(UnaryInstruction);}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitVACopyInst ( VACopyInst &  I )

inlineinherited

Definition at line 215 of file InstVisitor.h.

{ DELEGATE(IntrinsicInst); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitVAEndInst ( VAEndInst &  I )

inlineinherited

Definition at line 214 of file InstVisitor.h.

{ DELEGATE(IntrinsicInst); }

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitVAStartInst ( VAStartInst &  I )

inlineinherited

Definition at line 213 of file InstVisitor.h.

{ DELEGATE(IntrinsicInst); }

Instruction * InstCombiner::visitXor

(

BinaryOperator &

I

)

Definition at line 2221 of file InstCombineAndOrXor.cpp.

                                                     {
  bool Changed = SimplifyAssociativeOrCommutative(I);
  Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);

  if (Value *V = SimplifyVectorOp(I))
    return ReplaceInstUsesWith(I, V);

  if (Value *V = SimplifyXorInst(Op0, Op1, DL))
    return ReplaceInstUsesWith(I, V);

  // (A&B)^(A&C) -> A&(B^C) etc
  if (Value *V = SimplifyUsingDistributiveLaws(I))
    return ReplaceInstUsesWith(I, V);

  // See if we can simplify any instructions used by the instruction whose sole
  // purpose is to compute bits we don't care about.
  if (SimplifyDemandedInstructionBits(I))
    return &I;

  // Is this a ~ operation?
  if (Value *NotOp = dyn_castNotVal(&I)) {
    if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(NotOp)) {
      if (Op0I->getOpcode() == Instruction::And ||
          Op0I->getOpcode() == Instruction::Or) {
        // ~(~X & Y) --> (X | ~Y) - De Morgan's Law
        // ~(~X | Y) === (X & ~Y) - De Morgan's Law
        if (dyn_castNotVal(Op0I->getOperand(1)))
          Op0I->swapOperands();
        if (Value *Op0NotVal = dyn_castNotVal(Op0I->getOperand(0))) {
          Value *NotY =
            Builder->CreateNot(Op0I->getOperand(1),
                               Op0I->getOperand(1)->getName()+".not");
          if (Op0I->getOpcode() == Instruction::And)
            return BinaryOperator::CreateOr(Op0NotVal, NotY);
          return BinaryOperator::CreateAnd(Op0NotVal, NotY);
        }

        // ~(X & Y) --> (~X | ~Y) - De Morgan's Law
        // ~(X | Y) === (~X & ~Y) - De Morgan's Law
        if (isFreeToInvert(Op0I->getOperand(0)) &&
            isFreeToInvert(Op0I->getOperand(1))) {
          Value *NotX =
            Builder->CreateNot(Op0I->getOperand(0), "notlhs");
          Value *NotY =
            Builder->CreateNot(Op0I->getOperand(1), "notrhs");
          if (Op0I->getOpcode() == Instruction::And)
            return BinaryOperator::CreateOr(NotX, NotY);
          return BinaryOperator::CreateAnd(NotX, NotY);
        }

      } else if (Op0I->getOpcode() == Instruction::AShr) {
        // ~(~X >>s Y) --> (X >>s Y)
        if (Value *Op0NotVal = dyn_castNotVal(Op0I->getOperand(0)))
          return BinaryOperator::CreateAShr(Op0NotVal, Op0I->getOperand(1));
      }
    }
  }


  if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
    if (RHS->isOne() && Op0->hasOneUse())
      // xor (cmp A, B), true = not (cmp A, B) = !cmp A, B
      if (CmpInst *CI = dyn_cast<CmpInst>(Op0))
        return CmpInst::Create(CI->getOpcode(),
                               CI->getInversePredicate(),
                               CI->getOperand(0), CI->getOperand(1));

    // fold (xor(zext(cmp)), 1) and (xor(sext(cmp)), -1) to ext(!cmp).
    if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
      if (CmpInst *CI = dyn_cast<CmpInst>(Op0C->getOperand(0))) {
        if (CI->hasOneUse() && Op0C->hasOneUse()) {
          Instruction::CastOps Opcode = Op0C->getOpcode();
          if ((Opcode == Instruction::ZExt || Opcode == Instruction::SExt) &&
              (RHS == ConstantExpr::getCast(Opcode, Builder->getTrue(),
                                            Op0C->getDestTy()))) {
            CI->setPredicate(CI->getInversePredicate());
            return CastInst::Create(Opcode, CI, Op0C->getType());
          }
        }
      }
    }

    if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0)) {
      // ~(c-X) == X-c-1 == X+(-c-1)
      if (Op0I->getOpcode() == Instruction::Sub && RHS->isAllOnesValue())
        if (Constant *Op0I0C = dyn_cast<Constant>(Op0I->getOperand(0))) {
          Constant *NegOp0I0C = ConstantExpr::getNeg(Op0I0C);
          Constant *ConstantRHS = ConstantExpr::getSub(NegOp0I0C,
                                      ConstantInt::get(I.getType(), 1));
          return BinaryOperator::CreateAdd(Op0I->getOperand(1), ConstantRHS);
        }

      if (ConstantInt *Op0CI = dyn_cast<ConstantInt>(Op0I->getOperand(1))) {
        if (Op0I->getOpcode() == Instruction::Add) {
          // ~(X-c) --> (-c-1)-X
          if (RHS->isAllOnesValue()) {
            Constant *NegOp0CI = ConstantExpr::getNeg(Op0CI);
            return BinaryOperator::CreateSub(
                           ConstantExpr::getSub(NegOp0CI,
                                      ConstantInt::get(I.getType(), 1)),
                                      Op0I->getOperand(0));
          } else if (RHS->getValue().isSignBit()) {
            // (X + C) ^ signbit -> (X + C + signbit)
            Constant *C = Builder->getInt(RHS->getValue() + Op0CI->getValue());
            return BinaryOperator::CreateAdd(Op0I->getOperand(0), C);

          }
        } else if (Op0I->getOpcode() == Instruction::Or) {
          // (X|C1)^C2 -> X^(C1|C2) iff X&~C1 == 0
          if (MaskedValueIsZero(Op0I->getOperand(0), Op0CI->getValue())) {
            Constant *NewRHS = ConstantExpr::getOr(Op0CI, RHS);
            // Anything in both C1 and C2 is known to be zero, remove it from
            // NewRHS.
            Constant *CommonBits = ConstantExpr::getAnd(Op0CI, RHS);
            NewRHS = ConstantExpr::getAnd(NewRHS,
                                       ConstantExpr::getNot(CommonBits));
            Worklist.Add(Op0I);
            I.setOperand(0, Op0I->getOperand(0));
            I.setOperand(1, NewRHS);
            return &I;
          }
        } else if (Op0I->getOpcode() == Instruction::LShr) {
          // ((X^C1) >> C2) ^ C3 -> (X>>C2) ^ ((C1>>C2)^C3)
          // E1 = "X ^ C1"
          BinaryOperator *E1;
          ConstantInt *C1;
          if (Op0I->hasOneUse() &&
              (E1 = dyn_cast<BinaryOperator>(Op0I->getOperand(0))) &&
              E1->getOpcode() == Instruction::Xor &&
              (C1 = dyn_cast<ConstantInt>(E1->getOperand(1)))) {
            // fold (C1 >> C2) ^ C3
            ConstantInt *C2 = Op0CI, *C3 = RHS;
            APInt FoldConst = C1->getValue().lshr(C2->getValue());
            FoldConst ^= C3->getValue();
            // Prepare the two operands.
            Value *Opnd0 = Builder->CreateLShr(E1->getOperand(0), C2);
            Opnd0->takeName(Op0I);
            cast<Instruction>(Opnd0)->setDebugLoc(I.getDebugLoc());
            Value *FoldVal = ConstantInt::get(Opnd0->getType(), FoldConst);

            return BinaryOperator::CreateXor(Opnd0, FoldVal);
          }
        }
      }
    }

    // Try to fold constant and into select arguments.
    if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
      if (Instruction *R = FoldOpIntoSelect(I, SI))
        return R;
    if (isa<PHINode>(Op0))
      if (Instruction *NV = FoldOpIntoPhi(I))
        return NV;
  }

  BinaryOperator *Op1I = dyn_cast<BinaryOperator>(Op1);
  if (Op1I) {
    Value *A, *B;
    if (match(Op1I, m_Or(m_Value(A), m_Value(B)))) {
      if (A == Op0) {              // B^(B|A) == (A|B)^B
        Op1I->swapOperands();
        I.swapOperands();
        std::swap(Op0, Op1);
      } else if (B == Op0) {       // B^(A|B) == (A|B)^B
        I.swapOperands();     // Simplified below.
        std::swap(Op0, Op1);
      }
    } else if (match(Op1I, m_And(m_Value(A), m_Value(B))) &&
               Op1I->hasOneUse()){
      if (A == Op0) {                                      // A^(A&B) -> A^(B&A)
        Op1I->swapOperands();
        std::swap(A, B);
      }
      if (B == Op0) {                                      // A^(B&A) -> (B&A)^A
        I.swapOperands();     // Simplified below.
        std::swap(Op0, Op1);
      }
    }
  }

  BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0);
  if (Op0I) {
    Value *A, *B;
    if (match(Op0I, m_Or(m_Value(A), m_Value(B))) &&
        Op0I->hasOneUse()) {
      if (A == Op1)                                  // (B|A)^B == (A|B)^B
        std::swap(A, B);
      if (B == Op1)                                  // (A|B)^B == A & ~B
        return BinaryOperator::CreateAnd(A, Builder->CreateNot(Op1));
    } else if (match(Op0I, m_And(m_Value(A), m_Value(B))) &&
               Op0I->hasOneUse()){
      if (A == Op1)                                        // (A&B)^A -> (B&A)^A
        std::swap(A, B);
      if (B == Op1 &&                                      // (B&A)^A == ~B & A
          !isa<ConstantInt>(Op1)) {  // Canonical form is (B&C)^C
        return BinaryOperator::CreateAnd(Builder->CreateNot(A), Op1);
      }
    }
  }

  // (X >> Z) ^ (Y >> Z)  -> (X^Y) >> Z  for all shifts.
  if (Op0I && Op1I && Op0I->isShift() &&
      Op0I->getOpcode() == Op1I->getOpcode() &&
      Op0I->getOperand(1) == Op1I->getOperand(1) &&
      (Op0I->hasOneUse() || Op1I->hasOneUse())) {
    Value *NewOp =
      Builder->CreateXor(Op0I->getOperand(0), Op1I->getOperand(0),
                         Op0I->getName());
    return BinaryOperator::Create(Op1I->getOpcode(), NewOp,
                                  Op1I->getOperand(1));
  }

  if (Op0I && Op1I) {
    Value *A, *B, *C, *D;
    // (A & B)^(A | B) -> A ^ B
    if (match(Op0I, m_And(m_Value(A), m_Value(B))) &&
        match(Op1I, m_Or(m_Value(C), m_Value(D)))) {
      if ((A == C && B == D) || (A == D && B == C))
        return BinaryOperator::CreateXor(A, B);
    }
    // (A | B)^(A & B) -> A ^ B
    if (match(Op0I, m_Or(m_Value(A), m_Value(B))) &&
        match(Op1I, m_And(m_Value(C), m_Value(D)))) {
      if ((A == C && B == D) || (A == D && B == C))
        return BinaryOperator::CreateXor(A, B);
    }
  }

  // (icmp1 A, B) ^ (icmp2 A, B) --> (icmp3 A, B)
  if (ICmpInst *RHS = dyn_cast<ICmpInst>(I.getOperand(1)))
    if (ICmpInst *LHS = dyn_cast<ICmpInst>(I.getOperand(0)))
      if (PredicatesFoldable(LHS->getPredicate(), RHS->getPredicate())) {
        if (LHS->getOperand(0) == RHS->getOperand(1) &&
            LHS->getOperand(1) == RHS->getOperand(0))
          LHS->swapOperands();
        if (LHS->getOperand(0) == RHS->getOperand(0) &&
            LHS->getOperand(1) == RHS->getOperand(1)) {
          Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
          unsigned Code = getICmpCode(LHS) ^ getICmpCode(RHS);
          bool isSigned = LHS->isSigned() || RHS->isSigned();
          return ReplaceInstUsesWith(I,
                               getNewICmpValue(isSigned, Code, Op0, Op1,
                                               Builder));
        }
      }

  // fold (xor (cast A), (cast B)) -> (cast (xor A, B))
  if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
    if (CastInst *Op1C = dyn_cast<CastInst>(Op1))
      if (Op0C->getOpcode() == Op1C->getOpcode()) { // same cast kind?
        Type *SrcTy = Op0C->getOperand(0)->getType();
        if (SrcTy == Op1C->getOperand(0)->getType() && SrcTy->isIntegerTy() &&
            // Only do this if the casts both really cause code to be generated.
            ShouldOptimizeCast(Op0C->getOpcode(), Op0C->getOperand(0),
                               I.getType()) &&
            ShouldOptimizeCast(Op1C->getOpcode(), Op1C->getOperand(0),
                               I.getType())) {
          Value *NewOp = Builder->CreateXor(Op0C->getOperand(0),
                                            Op1C->getOperand(0), I.getName());
          return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
        }
      }
  }

  return Changed ? &I : nullptr;
}

Instruction * InstCombiner::visitZExt

(

ZExtInst &

CI

)

Definition at line 760 of file InstCombineCasts.cpp.

                                                 {
  // If this zero extend is only used by a truncate, let the truncate be
  // eliminated before we try to optimize this zext.
  if (CI.hasOneUse() && isa<TruncInst>(CI.user_back()))
    return nullptr;

  // If one of the common conversion will work, do it.
  if (Instruction *Result = commonCastTransforms(CI))
    return Result;

  // See if we can simplify any instructions used by the input whose sole
  // purpose is to compute bits we don't care about.
  if (SimplifyDemandedInstructionBits(CI))
    return &CI;

  Value *Src = CI.getOperand(0);
  Type *SrcTy = Src->getType(), *DestTy = CI.getType();

  // Attempt to extend the entire input expression tree to the destination
  // type.   Only do this if the dest type is a simple type, don't convert the
  // expression tree to something weird like i93 unless the source is also
  // strange.
  unsigned BitsToClear;
  if ((DestTy->isVectorTy() || ShouldChangeType(SrcTy, DestTy)) &&
      CanEvaluateZExtd(Src, DestTy, BitsToClear)) {
    assert(BitsToClear < SrcTy->getScalarSizeInBits() &&
           "Unreasonable BitsToClear");

    // Okay, we can transform this!  Insert the new expression now.
    DEBUG(dbgs() << "ICE: EvaluateInDifferentType converting expression type"
          " to avoid zero extend: " << CI);
    Value *Res = EvaluateInDifferentType(Src, DestTy, false);
    assert(Res->getType() == DestTy);

    uint32_t SrcBitsKept = SrcTy->getScalarSizeInBits()-BitsToClear;
    uint32_t DestBitSize = DestTy->getScalarSizeInBits();

    // If the high bits are already filled with zeros, just replace this
    // cast with the result.
    if (MaskedValueIsZero(Res, APInt::getHighBitsSet(DestBitSize,
                                                     DestBitSize-SrcBitsKept)))
      return ReplaceInstUsesWith(CI, Res);

    // We need to emit an AND to clear the high bits.
    Constant *C = ConstantInt::get(Res->getType(),
                               APInt::getLowBitsSet(DestBitSize, SrcBitsKept));
    return BinaryOperator::CreateAnd(Res, C);
  }

  // If this is a TRUNC followed by a ZEXT then we are dealing with integral
  // types and if the sizes are just right we can convert this into a logical
  // 'and' which will be much cheaper than the pair of casts.
  if (TruncInst *CSrc = dyn_cast<TruncInst>(Src)) {   // A->B->C cast
    // TODO: Subsume this into EvaluateInDifferentType.

    // Get the sizes of the types involved.  We know that the intermediate type
    // will be smaller than A or C, but don't know the relation between A and C.
    Value *A = CSrc->getOperand(0);
    unsigned SrcSize = A->getType()->getScalarSizeInBits();
    unsigned MidSize = CSrc->getType()->getScalarSizeInBits();
    unsigned DstSize = CI.getType()->getScalarSizeInBits();
    // If we're actually extending zero bits, then if
    // SrcSize <  DstSize: zext(a & mask)
    // SrcSize == DstSize: a & mask
    // SrcSize  > DstSize: trunc(a) & mask
    if (SrcSize < DstSize) {
      APInt AndValue(APInt::getLowBitsSet(SrcSize, MidSize));
      Constant *AndConst = ConstantInt::get(A->getType(), AndValue);
      Value *And = Builder->CreateAnd(A, AndConst, CSrc->getName()+".mask");
      return new ZExtInst(And, CI.getType());
    }

    if (SrcSize == DstSize) {
      APInt AndValue(APInt::getLowBitsSet(SrcSize, MidSize));
      return BinaryOperator::CreateAnd(A, ConstantInt::get(A->getType(),
                                                           AndValue));
    }
    if (SrcSize > DstSize) {
      Value *Trunc = Builder->CreateTrunc(A, CI.getType());
      APInt AndValue(APInt::getLowBitsSet(DstSize, MidSize));
      return BinaryOperator::CreateAnd(Trunc,
                                       ConstantInt::get(Trunc->getType(),
                                                        AndValue));
    }
  }

  if (ICmpInst *ICI = dyn_cast<ICmpInst>(Src))
    return transformZExtICmp(ICI, CI);

  BinaryOperator *SrcI = dyn_cast<BinaryOperator>(Src);
  if (SrcI && SrcI->getOpcode() == Instruction::Or) {
    // zext (or icmp, icmp) --> or (zext icmp), (zext icmp) if at least one
    // of the (zext icmp) will be transformed.
    ICmpInst *LHS = dyn_cast<ICmpInst>(SrcI->getOperand(0));
    ICmpInst *RHS = dyn_cast<ICmpInst>(SrcI->getOperand(1));
    if (LHS && RHS && LHS->hasOneUse() && RHS->hasOneUse() &&
        (transformZExtICmp(LHS, CI, false) ||
         transformZExtICmp(RHS, CI, false))) {
      Value *LCast = Builder->CreateZExt(LHS, CI.getType(), LHS->getName());
      Value *RCast = Builder->CreateZExt(RHS, CI.getType(), RHS->getName());
      return BinaryOperator::Create(Instruction::Or, LCast, RCast);
    }
  }

  // zext(trunc(X) & C) -> (X & zext(C)).
  Constant *C;
  Value *X;
  if (SrcI &&
      match(SrcI, m_OneUse(m_And(m_Trunc(m_Value(X)), m_Constant(C)))) &&
      X->getType() == CI.getType())
    return BinaryOperator::CreateAnd(X, ConstantExpr::getZExt(C, CI.getType()));

  // zext((trunc(X) & C) ^ C) -> ((X & zext(C)) ^ zext(C)).
  Value *And;
  if (SrcI && match(SrcI, m_OneUse(m_Xor(m_Value(And), m_Constant(C)))) &&
      match(And, m_OneUse(m_And(m_Trunc(m_Value(X)), m_Specific(C)))) &&
      X->getType() == CI.getType()) {
    Constant *ZC = ConstantExpr::getZExt(C, CI.getType());
    return BinaryOperator::CreateXor(Builder->CreateAnd(X, ZC), ZC);
  }

  // zext (xor i1 X, true) to i32  --> xor (zext i1 X to i32), 1
  if (SrcI && SrcI->hasOneUse() &&
      SrcI->getType()->getScalarType()->isIntegerTy(1) &&
      match(SrcI, m_Not(m_Value(X))) && (!X->hasOneUse() || !isa<CmpInst>(X))) {
    Value *New = Builder->CreateZExt(X, CI.getType());
    return BinaryOperator::CreateXor(New, ConstantInt::get(CI.getType(), 1));
  }

  return nullptr;
}

Instruction * llvm::InstVisitor< InstCombiner , Instruction * >::visitZExtInst ( ZExtInst &  I )

inlineinherited

Definition at line 183 of file InstVisitor.h.

{ DELEGATE(CastInst);}

bool InstCombiner::WillNotOverflowSignedAdd ( Value *  LHS, Value *  RHS  )

private

WillNotOverflowSignedAdd - Return true if we can prove that: (sext (add LHS, RHS)) === (add (sext LHS), (sext RHS)) This basically requires proving that the add in the original type would not overflow to change the sign bit or have a carry out. TODO: Handle this for Vectors.

Definition at line 898 of file InstCombineAddSub.cpp.

                                                                  {
  // There are different heuristics we can use for this.  Here are some simple
  // ones.

  // If LHS and RHS each have at least two sign bits, the addition will look
  // like
  //
  // XX..... +
  // YY.....
  //
  // If the carry into the most significant position is 0, X and Y can't both
  // be 1 and therefore the carry out of the addition is also 0.
  //
  // If the carry into the most significant position is 1, X and Y can't both
  // be 0 and therefore the carry out of the addition is also 1.
  //
  // Since the carry into the most significant position is always equal to
  // the carry out of the addition, there is no signed overflow.
  if (ComputeNumSignBits(LHS) > 1 && ComputeNumSignBits(RHS) > 1)
    return true;

  if (IntegerType *IT = dyn_cast<IntegerType>(LHS->getType())) {
    int BitWidth = IT->getBitWidth();
    APInt LHSKnownZero(BitWidth, 0);
    APInt LHSKnownOne(BitWidth, 0);
    computeKnownBits(LHS, LHSKnownZero, LHSKnownOne);

    APInt RHSKnownZero(BitWidth, 0);
    APInt RHSKnownOne(BitWidth, 0);
    computeKnownBits(RHS, RHSKnownZero, RHSKnownOne);

    // Addition of two 2's compliment numbers having opposite signs will never
    // overflow.
    if ((LHSKnownOne[BitWidth - 1] && RHSKnownZero[BitWidth - 1]) ||
        (LHSKnownZero[BitWidth - 1] && RHSKnownOne[BitWidth - 1]))
      return true;

    // Check if carry bit of addition will not cause overflow.
    if (checkRippleForAdd(LHSKnownZero, RHSKnownZero))
      return true;
    if (checkRippleForAdd(RHSKnownZero, LHSKnownZero))
      return true;
  }
  return false;
}

bool InstCombiner::WillNotOverflowUnsignedAdd ( Value *  LHS, Value *  RHS  )

private

WillNotOverflowUnsignedAdd - Return true if we can prove that: (zext (add LHS, RHS)) === (add (zext LHS), (zext RHS))

Definition at line 946 of file InstCombineAddSub.cpp.

                                                                    {
  // There are different heuristics we can use for this. Here is a simple one.
  // If the sign bit of LHS and that of RHS are both zero, no unsigned wrap.
  bool LHSKnownNonNegative, LHSKnownNegative;
  bool RHSKnownNonNegative, RHSKnownNegative;
  ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, DL, 0);
  ComputeSignBit(RHS, RHSKnownNonNegative, RHSKnownNegative, DL, 0);
  if (LHSKnownNonNegative && RHSKnownNonNegative)
    return true;

  return false;
}

Field Documentation

BuilderTy* llvm::InstCombiner::Builder

Definition at line 102 of file InstCombine.h.

const DataLayout* llvm::InstCombiner::DL

private

Definition at line 89 of file InstCombine.h.

char InstCombiner::ID = 0

static

Definition at line 104 of file InstCombine.h.

bool llvm::InstCombiner::MadeIRChange

private

Definition at line 91 of file InstCombine.h.

bool llvm::InstCombiner::MinimizeSize

private

Definition at line 93 of file InstCombine.h.

LibCallSimplifier* llvm::InstCombiner::Simplifier

private

Definition at line 92 of file InstCombine.h.

TargetLibraryInfo* llvm::InstCombiner::TLI

private

Definition at line 90 of file InstCombine.h.

InstCombineWorklist llvm::InstCombiner::Worklist

Worklist - All of the instructions that need to be simplified.

Definition at line 97 of file InstCombine.h.

---

The documentation for this class was generated from the following files:

• InstCombine.h
• InstCombineAddSub.cpp
• InstCombineAndOrXor.cpp
• InstCombineCalls.cpp
• InstCombineCasts.cpp
• InstCombineCompares.cpp
• InstCombineLoadStoreAlloca.cpp
• InstCombineMulDivRem.cpp
• InstCombinePHI.cpp
• InstCombineSelect.cpp
• InstCombineShifts.cpp
• InstCombineSimplifyDemanded.cpp
• InstCombineVectorOps.cpp
• InstructionCombining.cpp